Manual code: MSU_pigs.R

Transcription

1 Manual code: MSU_pigs.R Authors: Jose Luis Gualdrón Duarte 1 and Juan Pedro Steibel,3 1 Departamento de Producción Animal, Facultad de Agronomía, UBA-CONICET, Buenos Aires, ARG Department of Animal Science, Michigan State University, East Lansing, Michigan, USA 3 Department of Fisheries and Wildlife, Michigan State University, East Lansing, Michigan, USA Introduction The R code MSU_pigs.R is based on the paper Gualdrón et al This code uses a gpdata object called pigmsu as input file, which contains different files: genotype file, map, pedigree, and phenotype file. For further details of construction and structure of gpdata objects please refer to synbreed package in R (Wimmer et al. [1]).This package was used for assembling of gpdata pigmsu. Description of Input Files Each slot in pigmsu contains the following information: Genotype file (pigmsu$geno) In this file, animals IDs are row names and SNP names are column names. The genotypes are expressed as allelic dosage, having elements equal to 0,1,, i.e. the count of the allele used as reference, or a decimal number in the interval [0, ] for imputed genotypes.: MARC MARC ASGA ALGA01185 MARC M1GA Map file (pigmsu$map) This file contains SNP names as row names and also, chromosome and physical position expressed in Megabases as columns. chr pos MARC ASGA ASGA Pedigree file (pigmsu$pedigree) The pedigree file contains five columns: animal ID - Sire ID - Dam ID - generation - sex (1=male, =female). ID Par1 Par gener sex

2 Phenotype file (pigmsu$pheno) The trait file contains: animal ID in row names and the trait in column name: bf10_13wk Running MSU_pigs.R: Load required packages, functions and input files First of all, code MSU_pigs.R allows to load Regress package (version , R packages []) which is used in model fitting and variance components estimation process, performed by using REML algorithm.; and also, to load Synbreed package for data visualization and analysis of gpdata "pigmsu". library(synbreed) library(regress) Then, input gpdata (pigmsu) and functions (Functions_codes.R)for MSU_pigs.R are uploaded: load("pigmsu") source("functions_codes.r") # Load functions Filter process by Minor Allele Frequency (MAF) Genotypes in the F haves a second editing process considering MAF < 0.01, as well as fixed SNP (MAF= 0.5). As a result, a filtered matrix "M" that contains information for SNP is created. ### ## Filter by Minor allele frequency (MAF) in F genotypes # ## See Section: "Methods-Genotyping and data editing" in the reference paper # ### ## Genotype Matrix # ge<-pigmsu$geno IDrowM<-rownames(ge) IDrowM<-as.numeric(IDrowM) genf<-(idrowm>1000)&(idrowm<6000) Mf<-ge[genf,] # Genotype matrix for F animals m1<-colsums(mf) m<-m1/(*nrow(mf)) # Vector with frequency of second allele in each marker ## Filter by Minor allele frequency (MAF) and SNP fixed in the F genotypes MAF<-0.01 #Note: the addition of the filter "m<0.99" is because some SNPs are codificated as the major allele frequency f1<-(m>maf)&(m!=0.5) &(m<0.99) SNPf<-as.matrix(m[f1]) ## New genotype "M" matrix (SNP 44055) with genotypes filtered by MAF### M1<-colnames(Mf)%in%rownames(SNPf) M<-ge[,M1] # Total SNP for F0-F1-F generations

3 Construction of Z and G matrices Following the approach of VanRaden [3], Z matrix for F animals (Znf) is calculated applying the function "zstandard". Then, matrix G (GTo) for the same animalsis calculated using the Z matrix obtained previously (Znf). ################################ ##1) Computation of G Matrix ## ################################ IDrow<-rownames(M) IDrow<-as.numeric(IDrow) ## Frequency of P="" in the F0 genof0<-(idrow>6000) # logical vector with TRUE in F0 ids gen0<-m[genof0,] # Genotype for F0 (1st column ID - Genotype) all_frq_f0<-colmeans(gen0,na.rm=t)/ ## Computation of Z matrix for F idgf<-as.numeric(rownames(m)) idxg<-(idgf>1000)&(idgf<6000) # index for F animals ID genf<-m[idxg,] # genotype for F animals Znf<-zstandard(genf,alfreq=all_frq_f0,procedure="heterogeneous") ## Computation of G matrix: G=ZZ' GTo<-Znf%*%t(Znf) Fixed effects matrix construction From object "pigmsu" are extracted: 1) the growth trait 13 week tenth rib backfat (mm) (bf10_13wk) and ) sex for animals in the F, to be used as response variable (y) and incidence matrix fixed effects (X) for the model respectively. ####################################### ## ) Input files for funtion "snpe" ## ####################################### ##Phenotype pheno<-pigmsu$pheno[,"bf10_13wk",1] indx<-match(rownames(znf),names(pheno)) pheno<-pheno[indx] ##sex indx<-match(rownames(znf),pigmsu$pedigree$id) sex<-pigmsu$pedigree$sex[indx] #model.matrix to create matrix X sex<-as.factor(sex) x<-model.matrix( ~ sex -1, contrasts.arg=list(sex=contrasts(sex, contrasts=f))) #sex ID psex<-cbind(pigmsu$pedigree$id,pigmsu$pedigree$sex) #Extract ID and sex for all animals (F0-F1 and F) idxsex<-psex[,1]%in%names(pheno) sexid<-as.numeric(pigmsu$pedigree$id[idxsex]) rownames(x)<-sexid colnames(x)<-c("female","male")

4 Estimation of variance components and breeding values Once incidence matrix X is constructed, "snpe" function allows to fit the model y= X + a+ e. This function takes as input the phenotype file (y), X matrix, and also Z and G matrices. Internally, variance components are estimated by REML using regress package (version R []). ## 3) Apply function "snpe" ## # return: ## # g_hat=snpe,a_hat,e_variance,a_variance,heredability,#iteratons,ginv ## trout_snpe<-snpe(pheno,x,znf,gto) As a result, the function "snpe" gives: llik: Estimate of the LogLikelihood for the model a_hat: Genome breeding values (GEBVs). E_variance: A_variance: Error variance Additive variance e ) A) Heritability: Heritability of the trait n_iter: Ginv: The inverse of G matrix Number of iterations to converge. Here, a_hat contains the random breeding values, such that vector such that e N I, and I is the identity matrix. (0, e ) a N(0, G ), and e is the random error A Estimation of SNP effects The marker effect, variance of the markers effects, and p-values are calculated for each marker or SNP. For this purpose, the function "GWA" is applied using elements obtained in function "snpe". ## 4) Apply function "GWA" ## # return: ## # beta=uh ("g_hat"), snp_variance=vsnp, pvalues=pvalue ## gwa_trait<-gwa(trout_snpe,x,znf) As a result, the function "GWA" gives: beta: Estimate of each marker (SNP) effect snp_variance: Variance of each marker (SNP) effect pvalues: p-value for each marker (SNP) Histogram of p-values and Manhattan plot Here, the histogram of p-values and the Manhattan plot are displayed. However, for the Manhattan plot is necessary the absolute marker (SNP) position or consecutive position. Then, the function "abmap" is applied.

5 ############################## ## 5) Map Absolute Position ## ############################## map<-as.matrix(pigmsu$map) # Read map map1<-rownames(map)%in%colnames(znf) map<-map[map1,] # Final SNP #### Final map ##### mapmsu1<-abmap(map) # Apply funtion ab idxc<--map[,1]%% # Color index # histogram of p-values hist(gwa_trait$pvalues) # Manhattan Plot threshold<-0.05/nrow(map) #pdf(file="manhattan_trait.pdf") #option to save the Manhattan plot in format ".pdf" in the current directory plot(mapmsu1,-log(gwa_trait$pvalues,10),pch=16,col=ifelse(idxc==,"red","blue"),abline(h=log(threshold,10),lwd=1.1,col="red"),xlab="absolute position Mb", ylab="-log10(p-value)") #dev.off() #option to save the Manhattan plot in format ".pdf" in the current directory Definition of candidate segments Candidate segments are defined by taking SNPs within one Mb upstream and one Mb downstream of the SNP with smallest p-value in each chromosome (see list "maxlogpv"). ########################## ## 4) Candidate Segment ## ########################## pvalue<-gwa_trait$pvalues logpv<- -log(pvalue,10) assoc<-as.data.frame(cbind(map,mapmsu1,pvalue,logpv)) colnames(assoc)<-c("chr", "pos_mb","abspos_mb","pvalue","logpv") #Extract max -logpvalues per chromosome result <- vector("list",18) for(i in 1:18){ pvch<-assoc[assoc$chr==i,] maxpv<-which.max(pvch$logpv) mpv<-pvch[maxpv,] result[[i]] <-mpv } # List "maxlogpv" with the highest -log(p-value) per chromosome maxlogpv<-do.call(rbind,result) ########################################### ## 4.1) Extract segments per chromosome ## ########################################### #model_result<- vector("list",nrow(maxlogpv)) model_result<- vector("list",) # for(i in 1:nrow(maxlogpv)){ for(i in 1:){ #Example!!! Chromosome 1 and psnp<-maxlogpv[i,3] phigh<-as.numeric(psnp)+1 plow<-as.numeric(psnp)-1 #Extract SNP in this range slxmap<-(assoc[,3]<=phigh)&(assoc[,3]>=plow) slxmap<-assoc[slxmap,]

6 Computation of Z and G matrices for candidate segments Using the markers or SNPs into the segment, a new Z ("Z1") and G ("G1") matrices are calculated, whereas genomic relationship matrix G was built using all remaining SNPs. Next, the model equal to: y = X + a1+ a+ e ("model"), where a 1 is the vector of random effects associated with those SNP a1 N 0, G 1 A and a is the vector of additive random effects located in the segment, such that 1 associated with all SNPs except those involved with a 1, such that N 0, A a G. The model is compared with the reduce model y = X + a+ e ("model1"), from results obtained across the regress R package []. ################################################################### ## 4.) compute matrix Z for snp selected (normalized separtely) ## ################################################################### # Create Gseg for the segement using the columns in Znf belong to SNP segment Zidg<-colnames(Znf)%in%rownames(slxmap) Zs<-Znf[,Zidg] ################## Extrac for the GTo G<-GTo-(Zs%*%t(Zs)) ### compute matrix Z and G for snp selected in the segmwnt (normalized separtely) Zidx<-colnames(genf)%in%rownames(slxmap) genoffil<-genf[,zidx] idxfrq<-names(all_frq_f0)%in%rownames(slxmap) # Select the frequencies for the SNP segement sum(idxfrq==true) fil_frq_f0<-all_frq_f0[idxfrq] # Extract the frequencies for the SNP segment # Matrices Z1 and G1 (use the selected markers) #Apply the function for the SNP segement Z1<-zstandard(genoffil,alfreq=fil_frq_f0,procedure="heterogeneous") G1<-Z1%*%t(Z1) # G1 for SNP selected ########################################## ## 4.3) compute models with "regress" ## ########################################## model1<-regress(pheno~x,~gto) # y=xb+a+e model<-regress(pheno~x,~g1+g) # y=xb+a1+a+e ###################################### ## 4.4) Performed models comparison ## ###################################### llik<-list(result=list(l0=model1,l1=model)) ##################################################### ## 4.5) Results from Regress for model1 and model ## ##################################################### model_result[[i]]<-sapply(llik,summary.ll.fit) # Results refer to Additional files in "Table 1" } comp_model<-do.call(rbind,model_result)

7 The function summary.ll.fit list the results obtained by regress R package [] for "model1" and "model", as follow: Loglikem: LogLikem1: LRT: LRTseg: vare1: vara1: vare: VarA: LogLikehood "model" LogLikehood "model1" Likehood Ratio Test for "model1" and "model" p-value for Likehood Ratio Test for the segment Error variance e ) of "model1" Additive variance Error variance Additive variance of "model1" A) e ) of "model" A) of "model" Varseg: Additive variance segment A ) of "model" 1 Varseg_pr: Proportion in % of the total variance explained by the segment. Results for function summary.ll.fit are stored in the object "comp_model" To cite this code use: Gualdrón Duarte JL, Cantet RJC, Bates RO, Ernst CW, Raney NE, Steibel JP. Rapid screening for phenotype-genotype associations by linear transforming genomic evaluations. BMC Bioinformatics 014. (Submitted) References 1. Wimmer V, Albrecht T, Auinger H-J, Schön C-C: synbreed: a framework for the analysis of genomic prediction data using R. Bioinformatics 01, 8: Clifford D, McCullagh P: The regress function. R News 006, 6: VanRaden PM: Efficient methods to compute genomic predictions. J. Dairy Sci. 008, 91: