==== Front
PLoS One
PLoS One
plos
plosone
PLoS ONE
1932-6203 Public Library of Science San Francisco, CA USA 

10.1371/journal.pone.0241848
PONE-D-20-12127
Research Article
Biology and Life Sciences
Genetics
Single Nucleotide Polymorphisms
Computer and Information Sciences
Information Technology
Text Mining
Biology and Life Sciences
Genetics
Genomics
Biology and Life Sciences
Agriculture
Animal Products
Meat
Biology and Life Sciences
Nutrition
Diet
Food
Meat
Medicine and Health Sciences
Nutrition
Diet
Food
Meat
Biology and Life Sciences
Biochemistry
Metabolism
Metabolic Processes
Biology and Life Sciences
Biochemistry
Lipids
Lipid Metabolism
Biology and Life Sciences
Biochemistry
Metabolism
Lipid Metabolism
Biology and Life Sciences
Genetics
Heredity
Linkage Disequilibrium
Biology and Life Sciences
Genetics
Genetic Loci
Quantitative Trait Loci
Genome-wide identification of major genes and genomic prediction using high-density and text-mined gene-based SNP panels in Hanwoo (Korean cattle)
Genomic prediction using high-density and text-mined gene-based SNP panels in Hanwoo (Korean cattle)Lee Hyo Jun SoftwareWriting – original draft1 https://orcid.org/0000-0002-6906-6468Chung Yoon Ji Writing – original draft1 Jang Sungbong Software2 Seo Dong Won Writing – review & editing1 Lee Hak Kyo Data curationMethodology3 Yoon Duhak Data curationMethodology4 Lim Dajeong Data curationMethodology5* https://orcid.org/0000-0003-1508-4887Lee Seung Hwan ConceptualizationWriting – review & editing1* 1 
Division of Animal and Dairy Science, Chungnam National University, Daejeon, Korea
2 
Department of Animal and Dairy Science, University of Georgia, Athens, GA, United States of America
3 
Department of Animal Biotechnology, Chonbuk National University, Jeonju, Korea
4 
Department of Animal Science, Kyungpook National University, Sangju, Korea
5 
Animal Genome & Bioinformatics, National Institute of Animal Science, Wanju, Korea
Zhao Shuhong Editor Huazhong Agriculture University, CHINA
Competing Interests: The authors have declared that no competing interests exist.

* E-mail: lim_dj@korea.kr (DL); slee46@cnu.ac.kr (SHL)
2 12 2020 
2020 
15 12 e024184826 4 2020 21 10 2020 © 2020 Lee et al2020Lee et alThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.It was hypothesized that single-nucleotide polymorphisms (SNPs) extracted from text-mined genes could be more tightly related to causal variant for each trait and that differentially weighting of this SNP panel in the GBLUP model could improve the performance of genomic prediction in cattle. Fitting two GRMs constructed by text-mined SNPs and SNPs except text-mined SNPs from 777k SNPs set (exp_777K) as different random effects showed better accuracy than fitting one GRM (Im_777K) for six traits (e.g. backfat thickness: + 0.002, eye muscle area: + 0.014, Warner–Bratzler Shear Force of semimembranosus and longissimus dorsi: + 0.024 and + 0.068, intramuscular fat content of semimembranosus and longissimus dorsi: + 0.008 and + 0.018). These results can suggest that attempts to incorporate text mining into genomic predictions seem valuable, and further study using text mining can be expected to present the significant results.

This study was funded by awards from the Molecular Breeding program (Grant no. PJ0131692020) of the Next Generation BIOGREEN21 project of the National Institute of Animal Science, RDA, Republic of Korea. Hyo Jun Lee was also partly supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No.2020-0-01441, Artificial Intelligence Convergence Research Center (Chungnam National University)). Data AvailabilityData cannot be shared publicly by the authors due to legal restrictions on data use agreement sharing restrictions. All the data-set used in this study was provided by BioGreen 21 Program (Molecular Breeding Program) of National Institute of Animal Science, RDA. The carcass traits can be obtained at public web site (https://mtrace.go.kr/). Request for Genotype and meat quality traits data can be made to Korea National Institute of Animal Science, Animal Genome & Bioinformatics Division (http://www.nias.go.kr/english/sub/boardHtml.do?boardId=depintro), Tae Hun Kim, PhD, Director of Animal Genome & Bioinformatics Division, (thkim63@korea.kr). All other relevant data are within the paper and its Supporting Information filesData Availability
Data cannot be shared publicly by the authors due to legal restrictions on data use agreement sharing restrictions. All the data-set used in this study was provided by BioGreen 21 Program (Molecular Breeding Program) of National Institute of Animal Science, RDA. The carcass traits can be obtained at public web site (https://mtrace.go.kr/). Request for Genotype and meat quality traits data can be made to Korea National Institute of Animal Science, Animal Genome & Bioinformatics Division (http://www.nias.go.kr/english/sub/boardHtml.do?boardId=depintro), Tae Hun Kim, PhD, Director of Animal Genome & Bioinformatics Division, (thkim63@korea.kr). All other relevant data are within the paper and its Supporting Information files
==== Body
Introduction
Genomic prediction, which is the first step in genomic selection, is a method for calculating genomic estimated breeding values (GEBVs) using large numbers of genetic markers, such as single-nucleotide polymorphism (SNP), covering the whole genome [1]. The genomic prediction methods that are currently applied to livestock populations use the extent of linkage disequilibrium between markers and quantitative trait loci (QTL) because high-density SNPs increase the chances of co-segregation of markers with causal mutations [2]. Genetic variation in quantitative traits could be influenced by large numbers of loci affecting any given trait with small to moderate effects. In some cases, however, there are loci with moderate to large effects due to relatively recently selected mutations [3–5]. It is difficult to capture recently selected causal mutations in genomic prediction because the linkage disequilibrium between these mutations and other markers is incomplete [6]. Therefore, it is necessary to understand the genetic processes and information related to quantitative or complex traits more fully, as well as linkage disequilibrium between causal variants and common SNPs, to increase the ability of genomic prediction models. Genomic best linear unbiased prediction (GBLUP) is a commonly used method that has been widely utilized for genomic prediction. The main assumption of the GBLUP method is that most SNPs have small effects with a normal distribution, regardless of prior biological information on the genetic architecture of the traits [7]. However, the effects of SNPs associated with quantitative traits are not always normally distributed and the effects may differ depending on the biological processes of the traits. For these reasons, it might be necessary to incorporate previous biological knowledge into the GBLUP method for more accurate genomic prediction. In previous studies, when selected SNP panels based on biological information were weighted differentially in the GBLUP method, higher prediction accuracy was obtained compared with the normal GBLUP [8, 9]. In addition, using causal genes or markers with prior biological knowledge resulted in much more accurate QTL discovery [10].

As mentioned in the paragraph above, it is necessary to understand the biological characteristics of complex traits from previous studies for more accurate genomic prediction. However, manually scanning previous studies to analyze biological information requires a lot of time and effort because there are many published studies in the field of animal science, and the number is expanding at an increasing rate. As of 2018, approximately 29 million papers were cited in PubMed, one of the most commonly used life science databases (https://www.nlm.nih.gov/bsd/licensee/baselinestats.html). In addition, the majority of published papers are composed of unstructured text, which is difficult to use for other studies. Therefore, it is important to use techniques to extract useful information from the textual data without spending a lot of time. Text mining is one technique for resolving this problem [11]. In the biomedical field, text mining has been used to assist studies in gene–disease associations and gene–gene associations, and to analyze clinical datasets to improve quality of health care [12–14]. In addition, text mining has been widely applied in various fields other than biomedicine, such as business and marketing [15]. However, in the field of animal breeding, studies using text mining are still rare. The application of text mining to genomic prediction could be an interesting approach to animal breeding studies. In this study, text mining was used to identify genes associated with carcass and meat quality traits, and these text-mined genes with biological information were used for genomic prediction. The hypothesis of this study was that SNPs extracted from text-mined genes could be in tighter linkage disequilibrium with causal variants for carcass and meat quality traits, and weighting this SNP panel differentially in the GBLUP model could improve the performance of genomic prediction in cattle.

Materials and methods
Dataset
Hanwoo (Korean cattle) populations
The Animal Care and Use Committee of the National Institute of Animal Science (NIAS), Rural Development Administration (RDA), South Korea, approved the experimental procedures, and appropriate animal health and welfare guidelines were followed. The Hanwoo were sourced from two different commercial populations based on different phenotype measurements. The first commercial population included 12,635 individuals (animals were born between 2013 and 2016 and samples were collected between 2017 and 2019) evaluated for carcass traits (CWT, EMA, and BF). The second population consisted of 1,039 steers evaluated for meat quality traits (Warner–Bratzler Shear Force [WBSF] and intramuscular fat content). The two populations were half-sibs derived from 339 sires for the first population and 82 sires for the second population, with unrelated dams. All animals of the two populations (n = 12,635, n = 1,039) were slaughtered at averages of 918 and 920 days, respectively. The carcass traits (n = 12,635) consisted of three traits. The carcass weight (CWT/kg), backfat thickness (BF/mm), and eye muscle area (EMA/cm2) were measured after a 24-hour chill at the junction of the 12th and 13th ribs. Meat quality traits (n = 1,039) were measured by evaluating two traits in two muscles. The WBSF values of the longissimus dorsi muscle (D_SF) and semimembranosus muscle (S_SF) were measured according to the method described by Wheeler et al. (2000) [16]. Briefly, beef steak 2.5 cm2 thick was kept in polyethylene bags for 48 hours postmortem. All of the bags were heated in a water bath at 80°C for 30 minutes, until the internal temperature of the steaks reached 70°C. The samples were stored at room temperature for 30 minutes prior to measurement. An Instron Universal WBSF testing machine (Instron Corporation, Canton, MA) with a crosshead speed of 200 mm/min and a 50-kg load cell was used to measure the WBSF. Each sample was divided into six representative cores with a diameter of 1.27 cm and parallel to the muscle fibers. The final phenotype of the WBSF was the mean of the maximum force required to shear each core sample. The intramuscular fat contents of the longissimus dorsi muscle (D_IMF) and semimembranosus muscle (S_IMF) were measured using the microwave solvent extraction method described by AOAC International [17].

Genotyping and quality control
The genomic DNA of each animal group was extracted from longissimus thoracis muscle samples using a DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA). DNA concentration and purity were determined using a NanoDrop 1000 (Thermo Fisher Scientific, Wilmington, DE). A total of 13,674 samples were genotyped using the Illumina Bovine SNP50 BeadChip and the 1,295 samples were genotyped additionally by the Illumina Bovine HD BeadChip to use as the reference population in imputation step. All animals’ 50K genotypes were imputed to a high density level (777K) using Minimac3 [18]. r2 < 0.6 SNPs were excluded in the imputations step and SNPs on the sex chromosomes were excluded from the analysis. SNP quality control for each group was performed using PLINK1.9 software [19] based on the following criteria: minor allele frequency < 0.001 for carcass traits group and < 0.01 for meat quality traits group; gene call rate < 0.1. In the carcass trait group, 23,415 SNPs were excluded by the above step, leaving 670,080 SNPs. In the meat quality trait group, 56,477 SNPs were excluded by this step, and 637,017 SNPs were used for the analysis. The imputed 777K SNPs of each group were annotated using the SnpEff program [20].

Text mining and gene ontology term analysis
Published papers related to CWT, WBSF, IMF, BF, and EMA were searched before text mining. The workflow of the text mining is shown in S1 Fig. First, all the texts in the abstracts of papers containing queries related to traits in their abstracts or titles were collected. This step was performed using functions in the RISmed package of the R statistical programming language [21]. Words consisting of only capital letters or numbers were extracted to filter out words that were accidentally the same as gene symbols (e.g., impact, pigs). Finally, only words matching the bovine gene symbols in the BioMart databases were selected for analysis. The gene symbols were obtained from the Bioconductor package BiomaRT, and btaurus_gene_ensembl was used as the dataset [22]. SNPs contained in text-mined genes (TMG) were then extracted from the imputed 777K SNPs. Furthermore, SNPs from the intergenic region of TMG were also extracted because the intergenic region often contains functionally important elements, such as promoters and enhancers. The above two types of SNPs were used as text-mined SNPs. In this study, three marker sets—the imputed 777K SNPs (Im_777K), the SNPs excluding the text-mined SNPs from imputed 777K SNPs (exp_777K), and the text-mined SNPs—were used in genomic prediction. The Bioconductor R package ‘clusterProfiler’ was used for Gene Ontology (GO) analysis to identify the biological process of TMG [23]. The −log10 adjusted P-value (P.adj) by the Bonferroni method was used to examine the significance in GO analysis. To visualize the differences between QTL regions obtained from Animal QTL DB [24] and text-mined regions, karyotypes were plotted using the Circos program [25].

Statistical analyses
Genome-wide association study (GWAS) using text-mined gene-based SNP panels
The phenotypic data on carcass and meat quality traits were pre-adjusted for fixed effects including growing sites, birth year, season, and slaughter age using a linear model implemented in R software 3.3.1 (R Foundation for Statistical Computing, Vienna, Austria). The adjusted phenotypes and text-mined SNP panel were subsequently used for GWAS under a linear mixed model. The linear mixed model can be written as:
 yc=μ1N+Dβ+g+e 
where yc is a vector of the corrected phenotype for N individuals; μ is the overall mean of the term and 1N is a vector of N ones; D is a vector of genotype of the candidate SNPs recorded as 0, 1, or 2; β is the additive effect of the candidate SNPs; g is a vector of random polygenic effects from the genetic relationship matrix (GRM) constructed by the Im_777K; and e is a vector of residuals. This model was computed by GCTA 1.26 [26]. The GRM for the polygenic effect (g) was constructed using all SNPs except those on the chromosome where the candidate SNP was located. The P-values were adjusted using the Bonferroni method to correct multiple hypotheses. The values calculated by dividing 0.05 by the number of text-mined SNPs were used as the thresholds for obtaining significant SNPs associated with the trait.

Genomic models for estimation and prediction
The three genomic models were used to estimate genetic and residual variances as well as to predict genomic estimated breeding values (GEBV) in models 1 to 3. The two types of GRM constructed by lm_777K and exp_777K were used for models 1 and 2, respectively. The equations can be written as:
 y=μ1N+Xb+gall+e(model 1)  
 y=μ1N+Xb+g−t+e(model 2)  
where y is the vector of the observed phenotype for N individuals. X is an incidence matrix for the fixed effects and b is the vector of fixed effects, which included growing site, birth month, birth year, slaughter month, slaughter year, and slaughter age as covariates for all traits. In addition, the carcass traits included slaughter place and sex, while the meat quality trait included farm information (the owner’s name of steers). In the two equations, gall is the N vector of the additive effects from the GRM with lm_777K for additive genetic effects, and g−t is the N vector of the additive effects from the GRM with exp_777K. The genetic and residual effects were assumed to be normally distributed, with mean as zero. The variances estimated by the above two models are given by:
 Var[galle]=[Gallσ2all00Iσ2E],Var[g−te]=[G−tσ2−t00Iσ2E] 
where Gall and G−t are GRMs with lm_777K and exp_777K, respectively; and I is an N*N identity matrix.

In model 3, two GRMs constructed by exp_777K and text-mined SNPs were jointly used to differentially weight the random effects. The model used can be written as:
 y=μ1N+Xb+g−t+gt+e(model 3)  
where y is the vector of phenotypic observations, and gt is the N vector of the additive effects from GRM with the text-mined SNPs. The genetic and residual effects were assumed to be normally distributed, with mean as zero. The variances estimated by model 3 are given by:
 Var[gtg−te]=[Gtσt2000G−tσ−t2000Iσ2E] 
where Gt is the GRM with the text-mined SNPs.

Variance component estimation and GBLUP
The variance components, σall2,σ−t2, and σt2, and heritability were estimated using an average information restricted maximum likelihood (AIREML) model by implementing the AIREMLF90 program in the BLUPF90 family [27]. The proportion of genomic variance explained by each model can be written as:
 h2=σall2σall2 + e(model 1)  
 h2=σ−t2σ−t2+e(model 2)  
 h2=σt2 + σ−t2σt2 + σ−t2+e(model 3)  

GEBVs were predicted using GBLUP methods and a 10-fold cross-validation scheme was used to evaluate the accuracy of the GEBVs. Samples were divided into 10 groups of equal size. Nine of these groups were used as the reference set and the other group was used as the validation set in each cross-validation. The GEBVs for the model 1 and model 2 were calculated using the following mixed model. The matrix for the model used can be written as:
 [bu^]=[X′XX′ZZ′XZ′Z+λG−1]−1[X′yZ′y] 
where u^ is the vector of the GEBVs distributed as g~(0,Gσg2); G is genomic relationship matrix for individuals; Z is a design matrix designed one column for each GEBV and one row for each phenotype (if an individual would have no phenotype, Z would have a column with zero’s only for this individual). λ is shrinkage value calculated by (σ2e /σ2g). The GEBV for the model 3 is calculated using two random effect linear mixed model followed by
 [bu^−tu^t]=[X′XX′ZX′ZZ′XZ′Z+λ−tG−t−1Z′ZZ′XZ′ZZ′Z+λtGt−1]−1[X′yZ′yZ′y] 

Where u^−t and u^t are vectors of GEBVs calculated by exp_777K and text-mined SNPs; G−t and Gt are GRMs with exp_777k and text-mined SNPs. The final GEBV of model 3 is the sum of the two GEBVs (u^−t+u^t). The GRM (G) is defined as
 G=MM′2∑pj(1−pj) 
where M contains genotypes adjusted by allele frequency and pj is the allele frequency for marker j [28]. All of these estimates were performed using BLUPF90 [27]. The accuracy of predicted breeding values was calculated as the Pearson’s correlation between the GEBVs and adjusted phenotypes (yc) of the validation set, and the equation can be represented by:
 Accuracy=r(GEBV,yc) 

Results
Text mining and gene ontology term analysis
The queries used to search the papers and a statistical summary of the text mining are shown in Table 1. Regarding number of searched articles, CWT ranked first with 1893 papers, followed with IMF, WBSF, BF, EMA with (1854, 1097, 602, 546), respectively. In the number of calling genes, IMF showed the largest number of genes with 576, although a similar number of papers with CWT were searched. Other traits were ranked in order of CWT, BF, EMA, WBSF with (288, 195, 167, 156). The 30 genes that appeared with highest frequency in text mining are shown in Table 2. The most matched gene to bovine gene symbols in each trait were (CWT: IGF1(36 times), WBSF: CAST(110 times), IMF: SCD(105 times), BF: MC4R(35 times), and EMA: MSTN(19 times)), respectively.

10.1371/journal.pone.0241848.t001Table 1 Summary statistics of text mining and SNP calling.
Trait	Article6	Gene7	SNP8	Used query9	
CWT1	1,893	288	17,662	carcass weight[TIAB] OR dressed weight[TIAB]	
WBSF2	1,097	156	6,143	Warner-Bratzler Shear Force [TIAB] OR cuttability [TIAB] OR meat tenderness [TIAB]	
IMF3	1,854	576	30,983	intramuscular fat [TIAB]	
BF4	602	195	9,335	back fat [TIAB]	
EMA5	546	167	12,371	eye muscle area [TIAB] OR ribeye [TIAB] OR rib eye [TIAB]	
CWT1: Carcass weight; SF2: Warner-Bratzler Shear Force; IMF3: intramuscular fatty acid content; BF4: Backfat thickness; EMA5: Eye muscle area; Article6: number of articles searched in PubMed; Gene7: number of mined genes from searched articles; SNP8: number of SNPs called from imputed 777K markers; Used query9: queries used to search articles in PubMed.

10.1371/journal.pone.0241848.t002Table 2 The 30 genes symbol that appeared with highest frequency in text mining.
Trait	Symbol	Freq	Trait	Symbol	Freq	Trait	Symbol	Freq	
CWT	IGF1	36	BF	MC4R	35	EMA	MSTN	19	
MSTN	28	SST	26	CAPN1	18	
MC4R	25	IGF1	24	ADIPOQ	15	
LPL	24	FTO	18	LEPR	15	
TNF	24	GAA	16	PPARGC1A	15	
BLM	20	SLA	15	DES	13	
CAPN1	19	FASN	14	POMC	13	
IGFBP2	19	IGF2	14	GHR	12	
MGA	19	BSG	11	LEP	12	
NCAPG	19	MGA	11	PIK3C3	11	
POMC	18	RBP4	10	SLA	11	
IGF2	17	UCP2	10	CAST	9	
GHR	16	SPR	9	GH1	9	
AFP	15	CSTB	8	IGF2	9	
CRH	13	FABP3	8	LRIT3	9	
DGAT1	13	IGFBP3	8	LCORL	8	
FASN	13	LSR	8	MC4R	8	
GAA	13	MAP2K6	8	RPE	8	
LCORL	13	MTTP	8	ANGPTL3	7	
TRH	13	SCD	8	CRH	7	
CAPN3	11	STAT6	8	FABP4	7	
CAST	11	TNF	8	GRP	7	
SCD	11	CTSL	7	MAP2K6	7	
ABHD5	10	EZH2	7	AGAP3	6	
ASL	10	IRS4	7	BPI	6	
GNAS	10	MARK4	7	ERG	6	
IGFBP3	10	QSOX1	7	IGF1	6	
IGFBP4	10	SLC13A5	7	ADRB3	5	
IRS1	10	TGFBR1	7	EMD	5	
STAT6	10	UCP3	7	ME1	5	
WBSF	CAST	110	IMF	SCD	105		
CAPN1	104	LPL	80	
CAPN3	19	FABP4	70	
KCNJ11	18	FAS	54	
NES	17	FABP3	52	
DNAJA1	16	FASN	52	
MSTN	14	LEPR	47	
ADAMTS4	11	PPARG	38	
DGAT1	11	DGAT1	36	
HSPB1	9	MC4R	36	
SCD	8	AFP	27	
TNNT3	8	MSC	26	
UCP3	8	CAST	25	
ANGPTL3	7	PRKAG3	23	
IGFBP2	7	FTO	22	
ADAMTS5	6	SREBF1	22	
CAPN2	6	CAPN1	20	
DLK1	6	MAT2B	19	
MYOD1	6	PLIN2	17	
PRKAG3	6	RYR1	16	
STAT6	6	KLF6	15	
UCP2	6	ACACA	14	
LEP	5	ADH1C	14	
MMP2	5	GPAM	14	
APP	4	IGF2	14	
FABP4	4	PDHB	14	
GEN1	4	PPARA	14	
IGF2	4	ASIP	13	
LOX	4	MSTN	13	
MAP3K5	4	VRTN	13	
In the results of Gene Ontology (GO) term analysis (Table 3), CWT, BF, EMA-related TMG showed significance relatedness with growth regulator and growth factor (“response to hormone”, “regulation of signaling receptor activity”, and “response to endogenous stimulus”, “response to peptide”). WBSF-related TMG were identified to be associated with organic acid (“carboxylic acid metabolic process”, “oxoacid metabolic process”, “monocarboxylic acid biosynthetic process”, “organic acid metabolic process”, “monocarboxylic acid metabolic process”). For IMF, the biological process terms with lipid synthesis and lipid metabolism were statistically significant (“regulation of lipid metabolic process”, “lipid metabolic process”, “fatty acid metabolic process”, “regulation of lipid biosynthetic process”). The karyotypes of the QTL regions registered in animal QTLDB, text-mined regions, and the intersection of the two regions are shown in Fig 1. The highest percentage of intersecting regions within the text-mined regions corresponded to regions of CWT-related TMG (36.3%), and the lowest corresponded to IMF regions (5.5%).

10.1371/journal.pone.0241848.g001Fig 1 The karyotype of QTL regions registered in QTLDB, text-mined region, and the intersection of both regions.
Each karyotype represents the region for the trait indicated above. Percentages in parentheses beside the trait names indicate the ratio of text-mined region within QTLDB region.

10.1371/journal.pone.0241848.t003Table 3 The top five significant biological processes for each trait.
Trait	GO_ID	Biological process	GeneRatio1	−log10P.adj2:	
CWT	GO:0009725	response to hormone	19.8%	9.5	
GO:0010469	regulation of signaling receptor activity	21.4%	8.2	
GO:0009719	response to endogenous stimulus	24.6%	7.5	
GO:0043066	negative regulation of apoptotic process	19.0%	6.9	
GO:0043069	negative regulation of programmed cell death	19.0%	6.7	
WBSF	GO:0019752	carboxylic acid metabolic process	21.7%	2.6	
GO:0043436	oxoacid metabolic process	21.7%	2.4	
GO:0072330	monocarboxylic acid biosynthetic process	12.0%	2.3	
GO:0006082	organic acid metabolic process	21.7%	2.3	
GO:0032787	monocarboxylic acid metabolic process	14.5%	1.8	
IMF	GO:0019216	regulation of lipid metabolic process	11.5%	12.7	
GO:0032787	monocarboxylic acid metabolic process	15.3%	12.2	
GO:0006629	lipid metabolic process	23.4%	11.5	
GO:0006631	fatty acid metabolic process	11.1%	9.4	
GO:0046890	regulation of lipid biosynthetic process	7.7%	9.3	
BF	GO:0009725	response to hormone	23.4%	9.6	
GO:0032868	response to insulin	12.8%	8.2	
GO:1901700	response to oxygen-containing compound	28.7%	8.1	
GO:0009719	response to endogenous stimulus	28.7%	8.0	
GO:0043434	response to peptide hormone	12.8%	5.7	
EMA	GO:1901652	response to peptide	14.1%	4.1	
GO:0032868	response to insulin	11.3%	4.0	
GO:0010243	response to organonitrogen compound	19.7%	4.0	
GO:0043434	response to peptide hormone	12.7%	3.6	
GO:0062013	positive regulation of small molecule metabolic process	9.9%	3.5	
GeneRatio1: gene calling rate, i.e., the ratio of genes involved in each biological process among entire set of text-mined genes; −log10P.adj2: −log10 P-value adjusted by the Bonferroni method.

Genome-wide association study (GWAS) with text-mined SNPs
The Manhattan plots for each trait are shown in Fig 2. The Bonferroni correction method was used for the significance test (0.05/number of SNPs) in the genome-wide association study, and the SnpEff annotation information was referenced for marker locations. Three significant clusters were found in CWT. The most significant markers at position 10710350 in chromosome 4 are involved in the intron region of CALCR gene (P = 10−29.6). In the genomic region of chromosomes 6 and 14, markers involved in the LCORL–SLIT2 (position: 39,932,557) and PLAG1–CHCHD7 (position: 25,015,640) intergenic regions showed the most significance (P = 10−40.2 and P = 10−105.3). There are four significant genomic regions in BF. The most significant marker on chromosome 22 is located at a downstream gene variant of PPARG (position: 57,362,666; P = 10−6.5). The other most significant markers in chromosomes 2, 13, and 23 clusters are located in INSIG2–EN1 (position: 70,895,063; P = 10−5.97), APCDD1L–VAPB (position: 58,449,824; P = 10−7.3), and BMP5–HMGCLL1 (position: 4,622,146; P = 10−16.9) intergenic region. Three clusters showed significance in EMA. The most significant markers in chromosomes 3, 6, and 14 are involved in the S100A10–THEM4 (position: 18,822,190; P = 10−6.2), LCORL–SLIT2 (position: 39,932,557; P = 10−12.5), and PLAG1–CHCHD7 (position: 25,015,640; P = 10−26.6). For meat quality traits, only one marker at position 98,540,675 on chromosome 7 showed significance for D_SF (P = 10−7.4), located in an intron variant of the CAST gene.

10.1371/journal.pone.0241848.g002Fig 2 Manhattan plots with results of genome-wide association study using text-mined SNPs for each trait.
The y-axis shows the −log10P-value of each SNP and the x-axis is the marker index. The green line is the Bonferroni-line representing 0.05/number of markers. The blue line is the suggestive-line representing 0.1/number of markers.

Variance component estimation
A statistical summary of the variance component estimation is shown in Table 4. In carcass traits, CWT showed the highest heritability (0.42) when Im_777K was used in the estimation. BF and EMA showed no difference in heritability between the three different estimation models (BF: 0.41, EMA: 0.39). In meat quality traits, the heritabilities of WBSF in the two muscle types semimembranosus and longissimus dorsi were 0.1 and 0.19, respectively, when estimated using the Im_777K panel. S_IMF and D_IMF showed heritabilities of 0.21 and 0.32, respectively, when estimated using Im_777K. All four traits showed similar heritabilities between the three models.

10.1371/journal.pone.0241848.t004Table 4 Variance components at different marker set.
Trait	Value	Im_777K1	exp_777K2	exp_777K + tm_SNPs3	
CWT	σ2u	913.66	908.35	705.05 + 171.76	
σ2e	1287.6	1297.2	1307.3	
h2	0.42	0.41	0.4	
BF	σ2u	9.51	9.44	8.91 + 0.63	
σ2e	13.65	13.71	13.64	
h2	0.41	0.41	0.41	
EMA	σ2u	50.37	50.04	48 + 2.43	
σ2e	77.59	77.87	77.55	
h2	0.39	0.39	0.39	
S_SF	σ2u	0.11	0.11	0.07 + 0.04	
σ2e	1.02	1.02	1.02	
h2	0.1	0.09	0.09	
D_SF	σ2u	0.13	0.12	0.07 + 0.04	
σ2e	0.55	0.55	0.55	
h2	0.19	0.18	0.17	
S_IMF	σ2u	0.66	0.67	0.65+ 0.000024	
σ2e	2.46	2.44	2.47	
h2	0.21	0.22	0.21	
D_IMF	σ2u	5.28	5.24	4.34 + 0.73	
σ2e	11.51	11.55	11.72	
h2	0.32	0.31	0.3	
Im_777K1: estimated variance components with imputed 777K SNPs; exp_777K2: estimated variance components with imputed 777K SNPs except text-mined SNPs; exp_777K + tm_SNPs3: estimated variance components when using two marker sets (exp_777K, text-mined SNPs) to different genetic variance. First genetic variance was a component of exp_777K and second was a component of text-mined SNPs.

Genomic prediction
The accuracy of GEBV are shown separately for the carcass traits (CWT, BF, EMA) and meat quality traits (WBSF, IMF) in Table 5. Fitting two different GRMs constructed with two different SNP panels (exp_777K + tm_SNPs) as random effects in the GBLUP model showed better accuracy than fitting one GRM with exp_777K in all traits. In CWT, the prediction accuracy with Im_777K was 0.453, which was 0.002 higher than in the model with exp_777K + tm_SNPs. Conversely, for BF, using exp_777K + tm_SNPs resulted in an accuracy of 0.421, which was 0.002 higher than that using Im_777K. EMA also exhibited its highest prediction accuracy (0.437) when using two GRMs with exp_777K + tm_SNPs. The accuracy of genomic prediction using two GRMs for WBSF in the two muscle types, semimembranosus and longissimus dorsi, were calculated as 0.129 and 0.189, respectively, and those for IMF were 0.168 and 0.225, respectively, which were better than those using Im_777K. In order to validate the effect of text-mined SNPs in the multi-GRM model, GBLUP using evenly-mined SNPs (em_SNPs) and except SNPs was additionally conducted (Table 6). For all four meat quality traits, the GBLUP using tm_SNPs showed higher accuracy than em_SNPs. It seems that CWT and EMA may have more polygenic characteristics than other traits, because em_SNPs showed higher accuracy than tm_SNPs in these two traits.

10.1371/journal.pone.0241848.t005Table 5 Carcass traits average correlation between the GEBV and corrected phenotypic values (yc) and standard error for 10-validation set.
Meat quality traits average correlation between the GEBV and corrected phenotypic values (yc) and standard error for 10-validation set.

Trait	Im_777K	exp_777K	exp_777K + tm_SNPs	
CWT	0.453 ± 0.01	0.449 ± 0.01	0.451 ± 0.01	
BF	0.419 ± 0.01	0.413 ± 0.01	0.421 ± 0.01	
EMA	0.423 ± 0.01	0.429 ± 0.01	0.437 ± 0.004	
S_SF	0.105 ± 0.04	0.102 ± 0.02	0.129 ± 0.03	
D_SF	0.121 ± 0.03	0.115 ± 0.04	0.189 ± 0.03	
S_IMF	0.16 ± 0.02	0.15 ± 0.03	0.168 ± 0.02	
D_IMF	0.207 ± 0.04	0.163 ± 0.03	0.225 ± 0.02	
10.1371/journal.pone.0241848.t006Table 6 Accuracy of evenly-mined GBLUP and text-mined GBLUP.
Traits	exp_777k + tm_SNPs	exp_777k + em_SNPs1	
CWT	0.451 ± 0.01	0.471 ± 0.01	
BF	0.421 ± 0.01	0.419 ± 0.01	
EMA	0.437 ± 0.004	0.438 ± 0.01	
S_SF	0.129 ± 0.03	0.099 ± 0.02	
D_SF	0.189 ± 0.03	0.095 ± 0.02	
S_IMF	0.168 ± 0.02	0.147 ± 0.02	
D_IMF	0.225 ± 0.02	0.202 ± 0.03	
exp_777k + em_SNPs1: multi-GRM GBLUP with evenly-mined SNPs and except SNPs.

Discussion
Biological relatedness of text-mined gene with carcass and meat quality traits
Carcass traits
The top three mined genes for carcass traits were IGF1, MSTN, MC4R, SST, CAPN1, and PPARGC1A. Many previous studies have investigated the biological effect of these genes on the quantitative traits. Insulin-like growth factor (IGF) plays a key role in cell differentiation, growth, and metabolism regulation [29]. The myostatin (MSTN) gene, also known as GDF8, encodes a member of the transforming growth factor β superfamily, which is associated with the proper regulation of skeletal muscle mass and carcass yield in cattle [30]. The melanocortin 4 receptor (MC4R) gene plays an important role in energy balance and is associated with beef economic traits [31]. Peroxisome proliferator activated receptor gamma coactivator 1 alpha (PPARGC1A) have been standing out as a candidate gene for beef fat synthesis [32]. Although somatostatin (SST) inhibits growth hormone, there has been little research on the association between the SST gene and carcass traits. This gene seemed to have been mined because the abbreviation “SST” was used with other meanings, such as “sole soft tissue”, in the literature.

In addition to these high ranked genes, other genes (i.e., NCAPG, POMC, LCORL, FTO, IGF2, FABP3, LEPR, and ADIPOQ) were also found to be associated with growth-related traits in multiple breed [33–40]. The significant genes in GWAS results (CALCR, PLAG1, INSIG2, PPARG, BMP5, S100A10) also have been identified to have relationship with growth performance and obesity of adipose tissue for pig and cattle [35, 41–45]. In addition, many other TMG also seems to be associated with growth related traits because the GO term results revealed that carcass traits-related TMG were associated with growth regulator and growth factor.

Meat quality traits
The CAST and CAPN1 were included in the two most frequently mined genes related to the WBSF. Calpain 1 (CAPN1) encodes the large subunit of calcium-activated neutral proteases (calpain), and the calpastatin (CAST) gene inhibits μ- and m-calpain activity. These two proteins, as key myofibrillar proteins, mediate proteolysis during postmortem storage of the carcass and cuts of meat at refrigerated temperatures and play important roles in meat tenderness [46]. The association between these CAST/CAPN1 and WBSF has been studied extensively [47–50]. In IMF, SCD, LPL, and FABP4 were the three most frequently mined genes. The stearoyl-CoA desaturase (SCD) gene encodes an enzyme involved in fatty acid biosynthesis, primarily the synthesis of oleic acid [51]. The lipoprotein lipase (LPL) gene encodes lipoprotein lipase, which provides triglyceride-derived fatty acids to adipose tissue [52]. Fatty-acid-binding protein 4 (FABP4) plays a number of important roles, including fatty acid uptake, transport, and metabolism in the muscle [53].

In addition to these genes, CAPN3, KCNJ11, DNAJA1 are also known to be associated with beef tenderness [54–56] and FABP3, LEPR, FASN, DGAT1 were reported to associated with IMF in previous studies [57–59]. In the results of GO term analysis for WBSF, biological processes related to the carboxylic acid biosynthetic and metabolic processes were significant. Carboxylic acid is an organic acid that was shown in previous studies to affect beef tenderness [60, 61]. In addition, IMF related TMG showed a significant association with the regulation of lipid metabolic and biosynthetic processes. According to these biological processes, GO term results can support that WBSF, IMF–related TMG have been associated with WBSF and IMF.

Genomic prediction
When excluding text-mined SNPs from the Im_777K marker panels, the prediction accuracy for CWT, BF, WBSF, and IMF were decreased. In a previous simulation study, a panel that excluded QTL from the 50K SNP panel showed lower accuracy than a panel that included the QTL [2]. These results indicated that text-mined SNPs may be more strongly functionally associated with QTL for CWT, BF, WBSF, and IMF and include markers in a linkage disequilibrium relationship with QTL for these traits. Fitting two GRMs constructed using exp_777K and text-mined SNPs in the GBLUP model as different random effects resulted in higher accuracy than fitting one GRM constructed using Im_777K for BF, EMA, WBSF, and IMF. These results were consistent with previous studies indicating that differentially weighted subsets of markers based on genomic features increased the predictive ability [8]. The increase in accuracy was greater in the traits related to the longissimus dorsi muscle than in those related to the semimembranosus muscle. One of the most important factors that can affect the accuracy of genomic prediction is linkage disequilibrium between common SNPs and QTL [7]. As selection for a specific trait proceeds, linkage disequilibrium between causal polymorphisms for that trait and other marker loci appears to be stronger [6]. As traits related to the semimembranosus muscle were not considered in evaluating the degree of the Hanwoo breed, the selection of these traits would not have been carried out actively. Therefore, linkage disequilibrium between QTL and other markers would be weakened, and this seemed to have been responsible for these results.

In this study, the SNPs that seemed to be related to the traits were selected by text mining, and the prediction accuracy was slightly increased when these SNPs were weighted differentially to other SNP panels. In the GBLUP method, the weights of GRMs are controlled by the lambda value (σ2e /σ2u). As σ2u estimated by text-mined SNPs showed lower variance than estimated by exp_777K, higher lambda values were multiplied to GRM made by text-mined SNPs and this seemed to increase the prediction accuracy by giving more weight to text-mined SNPs in the model. Nevertheless, in comparisons between multi-GRM models, the accuracy of CWT and EMA decreased when tm_SNPs was used. These results may indicate that text-mined GBLUP doesn’t seem to be effective in the case of traits that are more genetically affected by polygenic effect than causal variant effect. There may be limits to the conclusion that text mining can improve prediction accuracy, since text mined SNPs didn’t result in a significant improvement in prediction accuracy. However, there was a slight accuracy increase for meat quality traits and GO term analysis may suggests that text mining can play a role in finding functional genes for complex traits. Therefore, attempts to incorporate text mining into genomic predictions seem valuable and further study (i.e., other SNP effects weighting methods) using text mining can be expected to present the significant results [62, 63]. In addition, text mining may be used for various population or breeds, since marker selection by text mining didn’t use the phenotypic or genetic information of a specific population.

Conclusions
This study was performed to use text mining, to extract biological information from previous papers and increase the performance of genomic prediction. The results showed that text mining could be used to find genes related to specific traits because associations between each carcass and meat quality trait and TMG were identified in the results of text mining and GO term analysis. However, a word that was accidentally the same as a gene symbol but used with another meaning (i.e., SST) was also mined as a text-mined gene. Therefore, it will be necessary to develop further methods of text mining that can resolve this problem. In the genomic prediction results, text-mined SNPs seemed to be in tighter linkage disequilibrium with QTL for BF, EMA, WBSF, and IMF. There may be limits to the conclusion that text mining can improve prediction accuracy, since text mined SNPs didn’t result in a significant improvement in prediction accuracy. However, attempts to incorporate text mining into genomic predictions still seem valuable, and further study using text mining can be expected to present the significant results, because a slight accuracy increase for meat quality traits may suggests that text mining can play a role in finding functional genes for complex traits. In addition, text mining may be used for various population or breeds, since marker selection by text mining didn’t use the phenotypic or genetic information of a specific population.

Supporting information
S1 Fig The workflow of the text mining.
(TIF)

Click here for additional data file.

 S1 File SNP information used in this study.
(ZIP)

Click here for additional data file.

 This study was preformed to develop new genomic selection model for carcass traits. We acknowledgment to Korea Institute for Animal Products Quality Evaluation to provide tissue sample for Genomic Reference Population.

10.1371/journal.pone.0241848.r001
Decision Letter 0
Zhao Shuhong Academic Editor © 2020 Shuhong Zhao2020Shuhong ZhaoThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Submission Version0
3 Jun 2020


PONE-D-20-12127

Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle)

PLOS ONE

Dear Dr. Lee,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Jul 18 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.



If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Shuhong Zhao, Ph.D

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you are reporting an analysis of a microarray, next-generation sequencing, or deep sequencing data set. PLOS requires that authors comply with field-specific standards for preparation, recording, and deposition of data in repositories appropriate to their field. Please upload these data to a stable, public repository (such as ArrayExpress, Gene Expression Omnibus (GEO), DNA Data Bank of Japan (DDBJ), NCBI GenBank, NCBI Sequence Read Archive, or EMBL Nucleotide Sequence Database (ENA)). In your revised cover letter, please provide the relevant accession numbers that may be used to access these data. For a full list of recommended repositories, see http://journals.plos.org/plosone/s/data-availability#loc-omics or http://journals.plos.org/plosone/s/data-availability#loc-sequencing.

3. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

4. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: https://www.youtube.com/watch?v=_xcclfuvtxQ

5. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

Additional Editor Comments (if provided):

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. 

Reviewer #1: Yes

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously? 

Reviewer #1: N/A

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Lee et.al conducted both GWAS and GS in a Korean cattle population. Except for the regular information of genotype and phenotype for GWAS and GS, the reported SNPs that associated with six traits were also identified by a text mining technology and used in GWAS and GS. The GWAS was carried out by a regular mixed linear model and the GS was performed by a regular GBLUP model with a single random effect or two random effects. The idea is interesting and the experimental design is fine. However, the manuscript is not well written and it is hard to judge if the statistical analyzes are done correctly. Therefore, I have the following concerns.

Major concerns:

Combined with the information from Table 1 and Table 4, we can find that the advantages of model 3 are bigger than the other two models when the number of text-mined SNPs is smaller. To remove the effect of the number of text-mined SNPs, I suggest adding one more experiment. Take CWT trait as an example, please randomly select 17,662 SNPs from all SNPs as 'fake text-mined SNPs' and test the accuracy of model 3 again and added as the fourth column in Table 4;

The published papers, which used this Korean cattle population, should not be used to identify the text-mined SNPs. This should be declared;

The GWAS was only carried out on text-mined SNPs, how about the GWAS results if using all SNPs?

The description of the GWAS model is not rigorous. For example, if the GWAS was performed by GCTA, the D should be a single text-mined SNP instead of a genotype matrix of text-mined SNPs. Please correct it and double-check the method section;

The slaughter information was added to the genomic prediction model as one of the fixed effects. What is slaughter information? Also, meat quality traits were adjusted by farmer information, what is ‘farmer information’? Was the farmer information added as fixed effects or random effects in the adjusted model. All details of the adjusted model should be clearly described.

Minor concerns:

A comma should be added in a number with more than three digits;

Why the MAF thresholds were set differently for carcass traits and meat quality traits?

The number of significant digits retained should be kept the same in each table;

What is TIAB in Table 1?

It would be more straightforward if the top genes with corresponding frequencies were shown in a Table instead of Figure1;

Some of the Chromosome IDs in Manhattan plots were overlapped.

Reviewer #2: The manuscript studied the effect of text-mined SNPs on GWAS and genomic prediction in Hanwoo. A major concern I have is that only three scenarios were compared for genomic prediction, and more scenarios should be included. Some suggestions are: 1) tm_SNPs only 2) a random subset of k SNPs from Im_777K, where k is the number of markers in exp_777K 3)fit the random subset obtained in (2) and the remaining as the two components in model 3

Other comments:

The standard error in table 4-1 is very low. Can you verify the standard error?

Why fixed effects are pre-adjusted in GWAS but not in genomic prediction?

Will GWAS using text-mined SNPs give lower power? Please add more discussion.

Line 160: Is \\beta for one candidate SNP in each test? Please correct it in line 158.

Line 209: Please correct it.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

10.1371/journal.pone.0241848.r002
Author response to Decision Letter 0
Submission Version1
13 Jul 2020


Manuscript PONE-D-20-12127R1

Dear Dr. Anita Estes

we appreciate the time and effort that you dedicated to providing feedback on our manuscript entitled “Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle)". 

As you comment, we revised some information to comply with PLOS ONE's data sharing policy.

 1) The relevant data used in this study were named as ‘Hanwoo reference population’ (carcass trait population) and ‘Hanwoo meat quality population’. In addition, none of the authors of this study have privileges in accessing datasets. 

 2) We revised the contact for accessing the genotype and meat quality traits data to (http://www.nias.go.kr/english/sub/boardHtml.do?boardId=depintro, National Institute of Animal Science, Animal Genome & Bioinformatics Division, Tae Hun Kim, PhD, Director of Animal Genome & Bioinformatics Division, thkim63@korea.kr). Data Availability in manuscript was also revised.

 3) All SNPs position and rs-numbers used in this study were added in Supporting Information. Detailed information and accession number of each SNP can be found in Bos_taurus_UMD_3.1.1 genome information submitted in NCBI GeneBank (accession: GCA_000003055.5).

Thank you again for your efforts, and please contact us if you have any further requests.

Kind regards

SeungHwan Lee

==================================================

Manuscript PONE-D-20-12127

Dear Dr. Shuhong Zhao

Thank you for giving us the opportunity to submit a revised manuscript ‘Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle)’ to PLOS ONE. We appreciate you for careful and thorough reading of our manuscript and valuable comment to our manuscript.

As you comment, the reference citation and figure file naming were changed following the PLOS ONE style templates. In addition, captions for Supporting Information files were added at the end of our revised manuscript.

We totally agree that the data sharing is important for reproducibility, replication, and validation of research. However, some of our data cannot be shared due to legal restrictions on data sharing. All the data-set (Phenotypes and genotypes for carcass traits) used in this study was provided by BioGreen 21 Program (Molecular Breeding Program) of National Institute of Animal Science, RDA. The carcass traits can be obtained at public web site (https://mtrace.go.kr/). Request for Genotype and meat quality traits data can be made to Korea National Institute of Animal Science (https://www.nias.go.kr:3443/front/main.do) at (lim.dj@korea.kr). We apologize for our situation and detailed data availability statement was added at the revised manuscript.

Response to Reviewers

Reviewer #1: Lee et.al conducted both GWAS and GS in a Korean cattle population. Except for the regular information of genotype and phenotype for GWAS and GS, the reported SNPs that associated with six traits were also identified by a text mining technology and used in GWAS and GS. The GWAS was carried out by a regular mixed linear model and the GS was performed by a regular GBLUP model with a single random effect or two random effects. The idea is interesting and the experimental design is fine. However, the manuscript is not well written and it is hard to judge if the statistical analyzes are done correctly. Therefore, I have the following concerns. 

Response: we appreciate the time and effort that you dedicated to providing feedback on our manuscript. We have incorporated most of your suggestions. Your insightful comment made our papers more valuable. We written in red for a point-by-point response to your concerns. Thank you.

Major concerns:

Combined with the information from Table 1 and Table 4, we can find that the advantages of model 3 are bigger than the other two models when the number of text-mined SNPs is smaller. To remove the effect of the number of text-mined SNPs, I suggest adding one more experiment. Take CWT trait as an example, please randomly select 17,662 SNPs from all SNPs as 'fake text-mined SNPs' and test the accuracy of model 3 again and added as the fourth column in Table 4;

Response: We appreciate the reviewer’s feedback, we agree that the analysis using random sets should be added. However, in the model 3, it is important to identify how independent between called (text-mined SNPs) and non-called (exp SNPs) regions are. If the SNPs are evenly (or randomly) extracted, both SNP set (called, non-called) will exhibit similar patterns due to LD. In the linear model, Fitting two highly correlated GRMs will be less accurate or similar to fitting the GRM using the whole SNPs. Therefore, we thought that using the evenly extracted SNPs to model 3 will not be much different from the results we want to show in model 1. The table below shows the results of Model 3 using evenly extracted SNPs and text-mined SNPs. the second table is the correlation between GRMs constructed with each SNP set.

　 text_mined + excepted evenly_extracted + excepted Im_777K

CWT 0.451 ± 0.01 0.4707 ± 0.01 0.453 ± 0.01

BF 0.421 ± 0.01 0.4187 ± 0.01 0.419 ± 0.01

EMA 0.437 ± 0.004 0.4376 ± 0.01 0.423 ± 0.01

S_SF 0.129 ± 0.03 0.0992 ± 0.02 0.105 ± 0.04

D_SF 0.189 ± 0.03 0.0954 ± 0.02 0.121 ± 0.03

S_IMF 0.168 ± 0.02 0.1471 ± 0.02 0.16 ± 0.02

D_IMF 0.225 ± 0.02 0.202 ± 0.03 0.207 ± 0.04

 text_mined + excepted evenly_extracted + excepted

CWT 0.586 0.784

BF 0.576 0.596

EMA 0.578 0.642

WBSF 0.71 0.981

IMF 0.908 0.997

The published papers, which used this Korean cattle population, should not be used to identify the text-mined SNPs. This should be declared;

Response: Thank you for pointing this out. There was no published paper using the Carcass trait population (n = 12,635) when conducting text-mining (sep, 2019). There is one published paper using part of the Meat quality population (n = 1,039), however the abstract of this paper does not contain the Gene symbol. Therefore, we can declare that published papers using this Korean cattle dataset were not included in text-mining.

The GWAS was only carried out on text-mined SNPs, how about the GWAS results if using all SNPs? 

Response: Indeed, we had expected that GWAS with text-mined SNPs would be able to identify the distribution of QTLs whose effect disappeared during the selection process. However, these results were not much different with the all SNPs results. It seems that text-mined SNPs have not been able to filter out common SNPs that have little or no effect but appear to be high due to the LD, because the text-mined SNPs were distributed throughout the whole genomes. The figures below show the GWAS results using the all SNPs and the simulation results for the QTL effect chainging with selection.

G0 (no selection)

G10 (after 10 generation with selection)

The description of the GWAS model is not rigorous. For example, if the GWAS was performed by GCTA, the D should be a single text-mined SNP instead of a genotype matrix of text-mined SNPs. Please correct it and double-check the method section;

Response: We apologize for the confusion. In this study, single marker linear mixed model was used for GWAS. Therefore, D must be genotype vector for each SNP, not genotype matrix. We corrected GWAS description (line 160) to ‘D is a vector of genotype of the candidate SNPs recorded as 0,1, or2’. In addition, in single marker liner mixed model, beta of each model means additive effect of single marker. Therefore, we also revised description (line 161) to ‘beta is the additive effect of the candidate SNPs’ 

The slaughter information was added to the genomic prediction model as one of the fixed effects. What is slaughter information? Also, meat quality traits were adjusted by farmer information, what is ‘farmer information’? Was the farmer information added as fixed effects or random effects in the adjusted model. All details of the adjusted model should be clearly described.

Response: Thank you for pointing this confusion out. In this study, the fixed effect (birth information, slaughter information) used in common for both population (carcass, meat quality) mean birth month, birth year, slaughter month, and slaughter year. We have the slaughter place information for carcass traits population, therefore we used this information as fixed effect for carcass traits. In case of meat quality population, we use the farmer information as fixed information, because we had the name of the owner for the slaughter steer (farmer information). To clarify these, we rewrote (line 177) to ‘b is the vector of fixed effects, which included growing site, birth month, birth year, slaughter month, slaughter year, and slaughter age as covariates for all traits. In addition, the carcass traits included slaughter place and sex, while the meat quality trait included farmer information (the owner’s name of steers).’

Minor concerns:

A comma should be added in a number with more than three digits.

Response: we revised all number with more than three digits.

Why the MAF thresholds were set differently for carcass traits and meat quality traits?

Response: You have raised an important point, we wanted to remove SNPs that less than 10 individuals had, because SNPs that too few individuals had are at risk of overestimating effects in GWAS or Genomic prediction. Since 10 individuals were equivalent to 0.001 of the carcass population and 0.01 of the meat quality population, different MAF thresholds were set for two population.

The number of significant digits retained should be kept the same in each table

Response: The significant digits and Gene ratio shown in table2 (table3 in the revised manuscript) were modified to show only one decimal point.

What is TIAB in Table 1?

Response: [TIAB] is one of the search options for Pubmed search engine. With [TIAB] option, only the papers which included words and numbers in a citation’s title, collection title, abstract, other abstract and keywords were search in Pubmed database. Other options and descriptions for Pubmed engine can be found in ‘https://www.ncbi.nlm.nih.gov/books/NBK3827/’.

It would be more straightforward if the top genes with corresponding frequencies were shown in a Table instead of Figure1.

Response: Thanks for pointing out. As you commented, the contents of all Figure 1 have been changed to Table2 for more tidy result delivery.

Some of the Chromosome IDs in Manhattan plots were overlapped.

Response: The revised pictures for the overlapping Chromosome IDs were re-attached.

Reviewer #2: The manuscript studied the effect of text-mined SNPs on GWAS and genomic prediction in Hanwoo. A major concern I have is that only three scenarios were compared for genomic prediction, and more scenarios should be included. Some suggestions are: 1) tm_SNPs only 2) a random subset of k SNPs from Im_777K, where k is the number of markers in exp_777K 3)fit the random subset obtained in (2) and the remaining as the two components in model 3

Response: We appreciate the reviewer’s comment, we agree that the analysis using tm_SNPs only and random sets should be added. In scenarios using only tm_SNPs, prediction accuracy was lower than other scenarios. There seem to be two reason for these results. 1) there are some polygenic effect missing, i.e. interaction between markers, since only tm_SNP is used for genomic prediction. 2) text-mining didn’t identify all causal variant. We didn’t wrote these tm_SNPs results in the manuscript, because these results were thought to be inevitable in a genomic prediction using low density marker set. 

In addition, we had conducted the genomic prediction fitting evenly selected SNPs to model 3. However, the independence of the two GRM (tm_SNP + exp_SNP) is important for statistical power of model 3 because the collinearity of two GRMs could make model weaken. Therefore, we thought that using the evenly (or randomly) extracted SNPs to model 3 will not be much different from the results we want to show in model 1. The table below shows the results of Model 3 using evenly extracted SNPs and text-mined SNPs. the second table is the correlation between GRMs constructed with each SNP set.

 

　 text_mined + excepted evenly_extracted + excepted Im_777K

CWT 0.451 ± 0.01 0.4707 ± 0.01 0.453 ± 0.01

BF 0.421 ± 0.01 0.4187 ± 0.01 0.419 ± 0.01

EMA 0.437 ± 0.004 0.4376 ± 0.01 0.423 ± 0.01

S_SF 0.129 ± 0.03 0.0992 ± 0.02 0.105 ± 0.04

D_SF 0.189 ± 0.03 0.0954 ± 0.02 0.121 ± 0.03

S_IMF 0.168 ± 0.02 0.1471 ± 0.02 0.16 ± 0.02

D_IMF 0.225 ± 0.02 0.202 ± 0.03 0.207 ± 0.04

 text_mined + excepted evenly_extracted + excepted

CWT 0.586 0.784

BF 0.576 0.596

EMA 0.578 0.642

WBSF 0.71 0.981

IMF 0.908 0.997

Other comments:

The standard error in table 4-1 is very low. Can you verify the standard error?

Response: We calculated standard error by

1/√K sd{〖CV〗_1 (〖cor〗^(-(1) ) ),…〖CV〗_K (〖cor〗^(-(K)))}

Where, 〖CV〗_n (〖cor〗^(-(n) ) ) is Pearson's correlation of test set of Validation N. In this study, we use 10 – fold validation for calculating accuracy. Therefore, we calculated the standard deviation of {〖CV〗_1 (〖cor〗^(-(1) ) ),…〖CV〗_10 (〖cor〗^(-(10)))}, and the value divided by √10 (1/√10 sd{〖CV〗_1 (〖cor〗^(-(1) ) ),…〖CV〗_10 (〖cor〗^(-(10)))}) was taken as the standard error.

Why fixed effects are pre-adjusted in GWAS but not in genomic prediction?

Response: We apologize for the confusion. and thank you for pointing this problem out. The GCTA program conduct association study by two steps, in order to improve computational efficiency. step1) the phenotype was adjusted by the mean and covariates (fixed effect), step2) the adjusted phenotype will subsequently be used for testing SNP association. Therefore, we pre-adjusted the phenotype to skip the step1. On the other hand, we didn’t pre-adjusted phenotype in genomic prediction, because the BLUPF90 program fits all fixed effects in one model. 

If SNPs are correlated with the fixed effects, pre-adjusting the phenotype by the fixed effects will probably cause overestimation of SNP effect. However, in this study, the environmental effect was used as a fixed effect. In addition, since SNP has not been mutated during the lifetime, there could be no interaction with the environment. Therefore, there are no difference between the two method (pre-adjusted or not).

Will GWAS using text-mined SNPs give lower power? Please add more discussion.

Response: In the individual selection processes for breeding, it is difficult to identify the QTL by GWAS as the effect of QTL decreases. Therefore, we expected that text-mining could filter out common-SNPs whose effect is overestimated by LD with causal variants. However, the GWAS using text-mined SNPs did not seem to significantly differ from the results using 777k SNPs. these results seem to be because text-mined SNPs were distributed throughout the whole genomes. Nevertheless, GWAS using text-mined SNPs still seems valuable in terms of identifying the association between causal variants candidate SNPs and phenotypes. The figures below show the GWAS results using the 777k SNPs and the simulation results for the QTL effect chainging with selection.

G0 (no selection)

G10 (after 10 generation with selection)

Line 160: Is \\beta for one candidate SNP in each test? Please correct it in line 158.

Response: We apologize for the confusion. In this study, single marker linear mixed model was used for GWAS. Therefore, beta must be one value for each SNP, not vector of the effect. We revised GWAS description (line 161) to ‘beta is the additive effect of the candidate SNPs’. Thank you for pointing out this problem. 

Line 209: Please correct it.

Response: We apologize for the miswriting, and We thank the reviewer for this comment. We wanted to present the equation for calculating the GEBV of the reference and validation set in a mixed model. However, there were some miswriting (i.e G is inverse of the GRM, not GRM). therefore, we removed this equation and added two new equation in (line 209, 215). First equation is linear mixed model with single random effect for the model1 and model2, second equation is mixed model with two random effect for the model3. We thought this way will more clearly explain our prediction models.

Attachment Submitted filename: Response.docx

Click here for additional data file.

 10.1371/journal.pone.0241848.r003
Decision Letter 1
Zhao Shuhong Academic Editor © 2020 Shuhong Zhao2020Shuhong ZhaoThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Submission Version1
14 Aug 2020


PONE-D-20-12127R1

Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle)

PLOS ONE

Dear Dr. Lee,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Sept 15. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.



If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Shuhong Zhao, Ph.D

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. 

Reviewer #1: Yes

Reviewer #2: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously? 

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: (No Response)

Reviewer #2: No

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have followed my recommendations to improve the manuscript. I suggest to add the information of tables and figures in point by point response to the manuscript and discuss the newly added results as well.

Reviewer #2: In author's reply, authors mentioned a potential reason for the results that prediction using tm_SNP only gives lower accuracy is "text-mining didn’t identify all causal variant.". If this is the reason, GWAS using tm_SNPs may cause lower power because causal variants excluded from tm_SNPs are not tested. In terms of prediction, significant differences are rarely observed, and it looks like a significant difference is observed for CWT, but "text_mined + excepted" is worse than "evenly_extracted + excepted". I would recommend authors to run more cross validation to further verify their conclusion. A paired t test may also be considered.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

10.1371/journal.pone.0241848.r004
Author response to Decision Letter 1
Submission Version2
24 Aug 2020


Reviewer #1: The authors have followed my recommendations to improve the manuscript. I suggest to add the information of tables and figures in point by point response to the manuscript and discuss the newly added results as well.

Thank you for this suggestion. As suggested by the reviewer, we added the evenly_SNPs + except results at table 6. 

Table 6. Accuracy of evenly-mined GBLUP and text-mined GBLUP

Traits exp_777k + tm_SNPs exp_777k + em_SNPs1

CWT 0.451 ± 0.01 0.471 ± 0.01

BF 0.421 ± 0.01 0.419 ± 0.01

EMA 0.437 ± 0.004 0.438 ± 0.01

S_SF 0.129 ± 0.03 0.099 ± 0.02

D_SF 0.189 ± 0.03 0.095 ± 0.02

S_IMF 0.168 ± 0.02 0.147 ± 0.02

D_IMF 0.225 ± 0.02 0.202 ± 0.03

exp_777k + em_SNPs1: multi-GRM GBLUP with evenly-mined SNPs and except SNPs.

These results have shown, in CWT, em_SNPs can make more accurate prediction than tm_SNPs. As with the results using whole SNPs, The polygenic characteristic of traits seem to make these results. These results may indicate that text-mined GBLUP doesn’t seem to be effective in the case of traits that are more genetically affected by polygenic effect than causal variant effect. We also added this finding in the discussion (line 398). Thank you again for your comment.

Reviewer #2: In author's reply, authors mentioned a potential reason for the results that prediction using tm_SNP only gives lower accuracy is "text-mining didn’t identify all causal variant.". If this is the reason, GWAS using tm_SNPs may cause lower power because causal variants excluded from tm_SNPs are not tested. In terms of prediction, significant differences are rarely observed, and it looks like a significant difference is observed for CWT, but "text_mined + excepted" is worse than "evenly_extracted + excepted". I would recommend authors to run more cross validation to further verify their conclusion. A paired t test may also be considered.

Thank you for pointing this out. We totally agree that if text-mining can’t identify all causal variant, the GWAS using tm_SNPs may have lower power than using whole SNPs. However, the lower accuracy of tm_SNPs only seems to be more affected by the loss of polygenic effect (e.g gene x gene, gene x SNPs etc) than missing the causal variant. Even though the text-mining misses a small number of causal variants (variant recall will be decreased.), it still seems good way to filter out a large number of common SNPs (variant precision will be increased.!). According to this description, this method may have lower power to identifying the genetic architecture of target traits. However, considering the ability to remove the SNPs that make noise to GWAS with LD, it seems to be a good method for finding true causal variants of target traits.

In terms of the genomic prediction results, as you commented on, only CWT and D_SF showed significant differences (by t-test). The reason why em_SNPs showed higher accuracy in CWT is seem to be because CWT is more affected by polygenic effect than causal variant effect due to the characteristic of quantitative traits (or carcass trait). We added these results at table 6, and discussion for these results also added (line 398). Thank you again for the time and effort that you dedicated to providing feedback on our manuscript.

Table 6. Accuracy of evenly-mined GBLUP and text-mined GBLUP

Traits exp_777k + tm_SNPs exp_777k + em_SNPs1

CWT 0.451 ± 0.01 0.471 ± 0.01

BF 0.421 ± 0.01 0.419 ± 0.01

EMA 0.437 ± 0.004 0.438 ± 0.01

S_SF 0.129 ± 0.03 0.099 ± 0.02

D_SF 0.189 ± 0.03 0.095 ± 0.02

S_IMF 0.168 ± 0.02 0.147 ± 0.02

D_IMF 0.225 ± 0.02 0.202 ± 0.03

exp_777k + em_SNPs1: multi-GRM GBLUP with evenly-mined SNPs and except SNPs.

Attachment Submitted filename: Response_to_reviewer-Final.docx

Click here for additional data file.

 10.1371/journal.pone.0241848.r005
Decision Letter 2
Zhao Shuhong Academic Editor © 2020 Shuhong Zhao2020Shuhong ZhaoThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Submission Version2
30 Sep 2020


PONE-D-20-12127R2

Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle)

PLOS ONE

Dear Dr. Lee,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Oct. 15th. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.



If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Shuhong Zhao, Ph.D

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. 

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously? 

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The manuscript has been improved and my previous concerns were addressed. I have one more comment:

It is very useful but also very challenge to weight a set of genetic markers to improve prediction accuracy. A recent study developed a new method to improve the prediction accuracy by weighting the contributions of genetic markers in the prediction model (Lilin Yin et. al, Genome Biology, 2020). It should be nice if this could be discussed to compare with your strategy.

Reviewer #2: I have no further comments.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

10.1371/journal.pone.0241848.r006
Author response to Decision Letter 2
Submission Version3
4 Oct 2020


Reviewer #1: The manuscript has been improved and my previous concerns were addressed. I have one more comment:

It is very useful but also very challenge to weight a set of genetic markers to improve prediction accuracy. A recent study developed a new method to improve the prediction accuracy by weighting the contributions of genetic markers in the prediction model (Lilin Yin et. al, Genome Biology, 2020). It should be nice if this could be discussed to compare with your strategy.

Response: Thank you for introducing a good method. We also agreed that weighting genetic markers is a challenge and further study is required. Therefore, we revised line 405 included at discussion to "Therefore, attempts to incorporate text mining into genomic predictions seem valuable and further study (i.e., other SNP effects weighting methods) using text mining can be expected to present the significant results.". In addition, the paper you recommended and (Wang H et al., Genetics Research, 2012) paper were referenced at the end of the sentence.

Reviewer #2: I have no further comments.

Response: Thank you for the time and effort you made during the whole review process.

Attachment Submitted filename: Response_to_reviewer.docx

Click here for additional data file.

 10.1371/journal.pone.0241848.r007
Decision Letter 3
Zhao Shuhong Academic Editor © 2020 Shuhong Zhao2020Shuhong ZhaoThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Submission Version3
22 Oct 2020


Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle)

PONE-D-20-12127R3

Dear Dr. Lee,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Shuhong Zhao, Ph.D

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

10.1371/journal.pone.0241848.r008
Acceptance letter
Zhao Shuhong Academic Editor © 2020 Shuhong Zhao2020Shuhong ZhaoThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
29 Oct 2020


PONE-D-20-12127R3 

Genome-wide identification of major genes and genomic prediction using High-Density and Text-Mined Gene-Based SNP panels in Hanwoo (Korean cattle) 

Dear Dr. Lee:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. 

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org. 

Thank you for submitting your work to PLOS ONE and supporting open access. 

Kind regards, 

PLOS ONE Editorial Office Staff

on behalf of

Mrs. Shuhong Zhao 

Academic Editor

PLOS ONE
==== Refs
References
1 Meuwissen T. , Hayes B. , and Goddard M. 
2001 
Prediction of total genetic value using genome-wide dense marker maps
. Genetics . 157 :1819 –1829
. 11290733 
2 Kizilkaya K. , Fernado R. L. , and Garrick D. J. , 2010 , Genomic prediction of simulated multibreed and purebred performance using observed fifty thousand single nucleotide polymorphism genotypes
. J. Anim. Sci . 88 :544 –551
. 10.2527/jas.2009-2064 
19820059 
3 Lango Allen H , Estrada K , Lettre G , Berndt SI , Weedon MN , Rivadeneira F , et al
2010 
Hundreds of variants clustered in genomic loci and biological pathways affect human height
. Nature. ;467 :832 –8
. 10.1038/nature09410 
20881960 
4 Yang J , Manolio TA , Pasquale LR , Boerwinkle E , Caporaso N , Cunningham JM , et al
2011 
Genome partitioning of genetic variation for complex traits using common SNPs
. Nat Genet . 43 :519 –25
. 10.1038/ng.823 
21552263 
5 Kemper K. , Goddard M. 
Understanding and predicting complex traits: knowledge from cattle
. 2012 
Hum Mol Genet . 21 :R45 –51
. 10.1093/hmg/dds332 
22899652 
6 Lewontin R. C. 
1964 
The interaction of selection and linkage. I. General considerations; heterotic models
. Genetics . 49 :49 –67
. .17248194 
7 Habier D. , Fernando R. , and Dekkers J. C. M. 
2007 
The impact of genetic relationship information on genome-assisted breeding values
. Genetics . 177 : 2389 –2397
. 10.1534/genetics.107.081190 
18073436 
8 Edwards S. M. , Sorensen I. F. , Sarup P. , Mackay T. F. , and Sorensen P. 
2016 
Genomic Prediction for Quantitative Traits Is Improved by Mapping Variants to Gene Ontology Categories in Drosophila melanogaster
. Genetics . 203 :1871 –1883
. 10.1534/genetics.116.187161 
27235308 
9 Fang Lingzhao , Sahana Goutam , Ma Peipei , Su Guosheng , Yu Ying , Zhang Shengli , et al
2017 
Exploring the genetic architecture and improving genomic prediction accuracy for mastitis and milk production traits in dairy cattle by mapping variants to hepatic transcriptomic regions responsive to intra-mammary infection
. Genet Sel Evol . 49 :44 
10.1186/s12711-017-0319-0 
28499345 
10 MacLeod I. M. , Bowman P. J. , Vande Jagat C. J. , Haile-Mariam M. , Kemper K. E. , Chamberlain A. J. , et al
2016 
Exploiting biological priors and sequence variants enhances QTL discovery and genomic prediction of complex traits. 2016
. BMC genomics . 17 :144 
10.1186/s12864-016-2443-6 
26920147 
11 Hearst M. A. 
1997 
Text data mining: Issues, techniques, and the relationship to information access
. Proc. UW/MS workshop on data mining .
12 Pletscher-Frankild S. , Pallejà A. , Tsafou K. , Binder J. X. , and Jensen L. J. J. M. 
2015 
DISEASES: Text mining and data integration of disease–gene associations
. Methods 
74 :83 –89
. 10.1016/j.ymeth.2014.11.020 
25484339 
13 Kankar P. , Adak S. , Sarkar A. , Murari K. , and Sharma G. 
2002 
MedMeSH summarizer: text mining for gene clusters
. Proc. SIAM International Conference on Data Mining .
14 Delespierre T. , Denormandie P. , Bar-Hen A. , Josseran L. J. B.  m. i. 2017 
Empirical advances with text mining of electronic health records
. BMC medical informatics and Decision Making 
17 :127 
10.1186/s12911-017-0519-0 
28830417 
15 Gálvez R. H. , and Gravano A. 
2017 
Assessing the usefulness of online message board mining in automatic stock prediction systems
. J. computational. Sci .19 :43 –56
. 10.1016/j.jocs.2017.01.001 
16 Wheeler T. , Shackelford S. , and Koohmaraie M. 
2000 
Relationship of beef longissimus tenderness classes to tenderness of gluteus medius, semimembranosus, and biceps femoris
. J. Anim. Sci . 78 :2856 –2861
. 10.2527/2000.78112856x 
11063309 
17 AOAC . 1996 
Official methods of analysis . 15th ed. 
AOAC Int. , Washington, DC .
18 Das Sayantan , Forer Lukas , Sebastian Schönherr , Carlo Sidore , Adam E Locke , Alan Kwong , et al
2016 
Next-generation genotype imputation service and methods
. Nature Genetics . 48 :1284 
10.1038/ng.3656 
27571263 
19 Purcell Shaun , Neale Benjamin , Todd-Brown Kathe , Lori Thomas , Manuel A.R. Ferreira , David Bender , et al
2007 
PLINK: a tool set for whole-genome association and population-based linkage analyses
. Am J Hum Genet . 81 :559 –575
. 10.1086/519795 
17701901 
20 Cingolani P. , Platts A. , Wang L. L. , Coon M. , Nguyen T. , Wang L. , et al
2012 
A program for annotating and predicting the effects of single nucleotide polymorphisms
, SnpEff. Fly . 6 :80 –92
. 10.4161/fly.19695 
22728672 
21 Kovalchik Stephanie . 2017 
RISmed: Download Content from NCBI Databases.

R package version 2.1.7 . https://CRAN.R-project.org/package = RISmed.
22 Durinck Steffen , Moreau Yves , Kasprzyk Arek , Davis Sean , Bart De Moor Alvis Brazma 
et al
2005 
BioMart and Bioconductor: a powerful link between biological databases and microarray data analysis
. Bioinformatics 
21 :3439 –3440
. 10.1093/bioinformatics/bti525 
16082012 
23 Yu Guangchuang , Wang Li-Gen , Han Yanyan  and He Qing-Yu . clusterProfiler: an R package for comparing biological themes among gene clusters
. 2012 
OMICS: A Journal of Integrative Biology . 16 :284 –287
. 10.1089/omi.2011.0118 
22455463 
24 Hu Z.-L. , Park C. A. , and Reecy J. M. 
2019 
Building a livestock genetic and genomic information knowledgebase through integrative developments of Animal QTLdb and CorrDB
. Nucleic Acids Research , 47 :D701 –D710
. 10.1093/nar/gky1084 
30407520 
25 Krzywinski M. , Schein J. , Birol I. , Connors J. , Gascoyne R. , Horsman D. , et al
2009 
Circos: an information aesthetic for comparative genomics
. Genome Res . 19 :1639 –1645
. 10.1101/gr.092759.109 
19541911 
26 Yang J. , Lee S.H. , Goddard M.E. , and Visscher P.M. 
2011 
GCTA: a tool for Genome-wide Complex Trait Analysis
. Am J Hum Genet . 88 :76 –82
. 10.1016/j.ajhg.2010.11.011 
21167468 
27 Misztal I. , Tsuruta S. , Strabel T. , Auvray B. , Druet T. , & Lee D. 
2002 
BLUPF90 and related programs (BGF90)
. Proc. the 7th world congress on genetics applied to livestock production, Montpellier, Fraance . 19 –23
.
28 VanRaden P. M. 
2008 
Efficient methods to compute genomic predictions
. Journal of Dairy Science . 91 :4414 –4423
. 10.3168/jds.2007-0980 
18946147 
29 BAXTER R.C. , 1988 
The insulin-like growth factors and their binding proteins
. Comparative Biochemistry and Physiology B . 91 (2 ), 229 –235
. 10.1016/0305-0491(88)90137-x 
2461835 
30 McPherron A. C. , and Lee S. J. 
1997 
Double muscling in cattle due to mutations in the myostatin gene
. PNAS . 94 :12457 –12461
. 10.1073/pnas.94.23.12457 
9356471 
31 Benoit S. , Schwartz M. , Baskin D. , Woods S. C. , and Seeley R. J. 
2000 
CNS melanocortin system involvement in the regulation of food intake
. Hormones and Behavior . 37 :299 –305
. 10.1006/hbeh.2000.1588 
10860674 
32 Samulin J. , Berget I. , Lien S. , Sundvold H. 
2008 
Differential gene expression of fatty acid binding proteins during porcine adipogenesis
. Comp. Biochem. Physiol. B Biochem. Mol. Biol. , 151 :147 –152
. 10.1016/j.cbpb.2008.06.010 
18621139 
33 Buchanan F. , Thue T. , Yu P. , and Winkelman‐Sim D. 
2005 
Single nucleotide polymorphisms in the corticotrophin‐releasing hormone and pro‐opiomelancortin genes are associated with growth and carcass yield in beef cattle
. Animal Genetics , 36 :127 –131
. 10.1111/j.1365-2052.2005.01255.x 
15771721 
34 Lindholm-Perry A. K. , Sexten A. K. , Kuehn L. A. , Smith T. P. , King D. A. , Shackelford S. D. , et al
2011 
Association, effects and validation of polymorphisms within the NCAPG-LCORL locus located on BTA6 with feed intake, gain, meat and carcass traits in beef cattle
. BMC Genetics . 12 :103 
10.1186/1471-2156-12-103 
22168586 
35 Nishimura Shota , Watanabe Toshio , Mizoshita Kazunori , Tatsuda Ken , Fujita Tatsuo , Watanabe Naoto , et al
2012 
Genome-wide association study identified three major QTL for carcass weight including the PLAG1-CHCHD7 QTN for stature in Japanese Black cattle
. BMC Genetics . 13 :40 
10.1186/1471-2156-13-40 
22607022 
36 Dina C. , Meyre D. , Gallina S. , Durand E. , Körner A. , Jacobson P. , et al
2007 
Variation in FTO contributes to childhood obesity and severe adult obesity
. Nature Genetics . 39 :724 –726
. 10.1038/ng2048 
17496892 
37 Cho S. A. , Park T. S. , Yoon D. H. , Cheong H. S. , Namgoong S. , Park B. L. , et al
2008 
Identification of genetic polymorphisms in FABP3 and FABP4 and putative association with back fat thickness in Korean native cattle
. BMB reports . 41 :29 –34
. 10.5483/bmbrep.2008.41.1.029 
18304447 
38 Vykoukalova Z. , Knoll A. , Dvořák J. , and Čepica S. 
2006 
New SNPs in the IGF2 gene and association between this gene and backfat thickness and lean meat content in Large White pigs
. Journal of Animal Breeding and Genetics . 123 :204 –207
. 10.1111/j.1439-0388.2006.00580.x 
16706926 
39 Hirose Kensuke , Ito Tetsuya , Fukawa Kazuo , Arakawa Aisaku , Mikawa Satoshi , Hayashi Yoichi , et al
2013 
Evaluation of effects of multiple candidate genes (LEP, LEPR, MC4R, PIK3C3, and VRTN) on production traits in Duroc pigs
. Animal Science Journal . 85 :3 
10.1111/asj.12134 
24128088 
40 Shin S. , and Chung E. 
2013 
Novel SNPs in the bovine ADIPOQ and PPARGC1A genes are associated with carcass traits in Hanwoo (Korean cattle)
. Mol. Biol. Rep . 40 :4651 –4660
. 10.1007/s11033-013-2560-0 
23649766 
41 Alexander L. S. , Qu A. , Cutler S. A. , Mahajan A. , Rothschild M. F. , Cai W. , et al
2010 
A calcitonin receptor (CALCR) single nucleotide polymorphism is associated with growth performance and bone integrity in response to dietary phosphorus deficiency
. J. Anim. Sci . 88 :1009 –1016
. 10.2527/jas.2008-1730 
19933433 
42 Grzes Maria , Sadkowski Slawomir , Rzewuska Katarzyna , Szydlowski Maciej  & Switonski Marek . 2016 
Pig fatness in relation to FASN and INSIG2 genes polymorphism and their transcript level
. Molecular Biology Reports . 43 :381 –389
, 10.1007/s11033-016-3969-z 
26965892 
43 Sevane N. , Armstrong E. , Cortés O. , Wiener P. , PongWong R. , Dunner S. , et al
2013 
Association of bovine meat quality traits with genes included in the PPARG and PPARGC1A networks
. Meat Science . 94 :328 –335
, 10.1016/j.meatsci.2013.02.014 
23567132 
44 Shao G.C. , Luo L.F. , Jiang S.W. , Deng C.Y. , Xiong Y.Z. , Li F.E. 
2011 
A C/T mutation in microRNA target sites in BMP5 gene is potentially associated with fatness in pigs
. Meat science . 87 :299 –303
. 10.1016/j.meatsci.2010.09.013 
21093991 
45 Kogelman Lisette J. A. , Zhernakova Daria V. , Westra Harm-Jan , Cirera Susanna , Fredholm Merete , Franke Lude , et al
2015 
An integrative systems genetics approach reveals potential causal genes and pathways related to obesity
. Genome Medicine . 7 :105 
10.1186/s13073-015-0229-0 
26482556 
46 Wheeler T. , and Koohmaraie M. 
1994 
Prerigor and postrigor changes in tenderness of ovine longissimus muscle
. J. Anim. Sci . 72 :1232 –1238
. 10.2527/1994.7251232x 
8056668 
47 Corva P. , Soria L. , Schor A. , Villarreal E. , Cenci M. P. , Motter M. , et al
2007 
Association of CAPN1 and CAST gene polymorphisms with meat tenderness in Bos taurus beef cattle from Argentina
. Genetics and Molecular Biology . 30 :1064 –1069
. 10.1590/S1415-47572007000600006 
48 Li Y. , Jin H. , Yan C. , Seo K. , Zhang L. , Ren C. , et al
2013 
Association of CAST gene polymorphisms with carcass and meat quality traits in Yanbian cattle of China
. Mol. Biol. Rep . 40 :1875 –1881
. 10.1007/s11033-012-2243-2 
23086304 
49 Schenkel F. , Miller S. , Jiang Z. , Mandell I. , Ye X. , Li H. , et al
2006 
Association of a single nucleotide polymorphism in the calpastatin gene with carcass and meat quality traits of beef cattle
. J. Anim. Sci . 84 : 291 –299
. 10.2527/2006.842291x 
16424255 
50 White S. N. , Casas E. , Wheeler T. L. , Shackelford S. D. , Koohmaraie M. , Riley D. G. , et al
2005 
A new single nucleotide polymorphism in CAPN1 extends the current tenderness marker test to include cattle of Bos indicus, Bos taurus, and crossbred descent
. J. Anim. Sci . 83 :2001 –2008
. 10.2527/2005.8392001x 
16100054 
51 Paton C. M. , Ntambi J. M. 
2009 
Biochemical and physiological function of stearoyl-CoA desaturase
. American Journal of Physiology-Endocrinology and Metabolism . 297 :E28 –E37
. 10.1152/ajpendo.90897.2008 
19066317 
52 Bonnet M. , Leroux C. , Faulconnier Y. , Hocquette J. F. , Bocquier F. , Martin P. , et al
2000 
Lipoprotein lipase activity and mRNA are up-regulated by refeeding in adipose tissue and cardiac muscle of sheep
. The Journal of Nutrition . 130 :749 –756
. 10.1093/jn/130.4.749 
10736325 
53 Kaikaus R. , Bass N. , and Ockner R. J. E. 
1990 
Functions of fatty acid binding proteins
. Experientia . 46 :617 –630
. 10.1007/BF01939701 
2193826 
54 Gandolfi G. , Pomponio L. , Ertbjerg P. , Karlsson A. H. , Costa L. , Lametsch R. , et al
2011 
Investigation on CAST, CAPN1 and CAPN3 porcine gene polymorphisms and expression in relation to post-mortem calpain activity in muscle and meat quality
. Meat science . 88 : 694 –700
. 10.1016/j.meatsci.2011.02.031 
21450414 
55 Malheiros J. M. , Enríquez-Valencia C. E. , da Silva Duran B. O. , de Paula T. G. , Curi R. A. , de Vasconcelos Silva , et al
2018 
Association of CAST2, HSP90AA1, DNAJA1 and HSPB1 genes with meat tenderness
 in Nellore cattle. Meat science . 138 , 49 –52
. 10.1016/j.meatsci.2018.01.003 
29331838 
56 Tizioto Polyana C. , Gasparin Gustavo , Souza Marcela M. , Mudadu Mauricio A. , Coutinho Luiz L. , Gerson B. Mourão , et al2013 
Identification of KCNJ11 as a functional candidate gene for bovine meat tenderness
. Physiological Genomics . 45 :1215 –1221
. 10.1152/physiolgenomics.00137.2012 
24151244 
57 Li X. , Kim S. W. , Choi J.S. , Lee Y.M. , Lee C.K. , Choi B.H. , et al
2010 
Investigation of porcine FABP3 and LEPR gene polymorphisms and mRNA expression for variation in intramuscular fat content
. Mol. Biol. Rep . 37 :3931 –3939
. 10.1007/s11033-010-0050-1 
20300864 
58 Abe T. , Saburi J. , Hasebe H. , Nakagawa T. , Misumi S. , Nade T. , et al
2009 
Novel mutations of the FASN gene and their effect on fatty acid composition in Japanese Black beef
. Biochem Genet . 47 :397 –411
. 10.1007/s10528-009-9235-5 
19291389 
59 Thaller G. , Kühn C. , Winter A. , Ewald G. , Bellmann O. , Wegner J. , et al
2003. DGAT1, a new positional and functional candidate gene for intramuscular fat deposition in cattle
. Animal Genetics . 34 :354 –357
. 10.1046/j.1365-2052.2003.01011.x 
14510671 
60 Aktaş N. , Aksu M. , and Kaya M. 
2003 
The effect of organic acid marination on tenderness, cooking loss and bound water content of beef
. J. Muscle Foods banner . 14 :181 –194
. 10.1111/j.1745-4573.2003.tb00699.x 
61 ARGANOSA G. C. , and MARRIOTT N. G. 
1989 
Organic acids as tenderizers of collagen in restructured beef
. J. Food Science . 54 :1173 –1176
. 10.1111/j.1365-2621.1989.tb05949.x 
62 Wang H. , Misztal I. , Aguilar I. , Legarra A. , and Muir W. M. 
2012 
Genome-wide association mapping including phenotypes from relatives without genotypes
. Genetics Research , 94 (2 ), 73 –83
. 10.1017/S0016672312000274 
22624567 
63 Yin L. , Zhang H. , Zhou X. , Yuan X. , Zhao S. , Li X. , et al
2020. KAML: improving genomic prediction accuracy of complex traits using machine learning determined parameters
. Genome biology , 21 (1 ), 1 –22
. 10.1186/s13059-019-1906-x 
31892341