Last updated: 2021-08-04
Checks: 7 0
Knit directory: implementGMSinCassava/
This reproducible R Markdown analysis was created with workflowr (version 1.6.2). The Checks tab describes the reproducibility checks that were applied when the results were created. The Past versions tab lists the development history.
Great! Since the R Markdown file has been committed to the Git repository, you know the exact version of the code that produced these results.
Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.
The command set.seed(20210504) was run prior to running the code in the R Markdown file. Setting a seed ensures that any results that rely on randomness, e.g. subsampling or permutations, are reproducible.
Great job! Recording the operating system, R version, and package versions is critical for reproducibility.
Nice! There were no cached chunks for this analysis, so you can be confident that you successfully produced the results during this run.
Great job! Using relative paths to the files within your workflowr project makes it easier to run your code on other machines.
Great! You are using Git for version control. Tracking code development and connecting the code version to the results is critical for reproducibility.
The results in this page were generated with repository version f6f58f0. See the Past versions tab to see a history of the changes made to the R Markdown and HTML files.
Note that you need to be careful to ensure that all relevant files for the analysis have been committed to Git prior to generating the results (you can use wflow_publish or wflow_git_commit). workflowr only checks the R Markdown file, but you know if there are other scripts or data files that it depends on. Below is the status of the Git repository when the results were generated:
Ignored files:
Ignored: .DS_Store
Ignored: .Rhistory
Ignored: .Rproj.user/
Ignored: analysis/.DS_Store
Ignored: analysis/figure/
Ignored: code/.DS_Store
Ignored: data/.DS_Store
Untracked files:
Untracked: analysis/speedUpPredCrossVar.Rmd
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.bed
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.bim
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.fam
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.hap.gz
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.log
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.nosex
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.sample
Untracked: data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719.vcf.gz
Untracked: data/DatabaseDownload_2021May04/
Untracked: data/blups_forCrossVal.rds
Untracked: data/config.txt
Untracked: data/config_mw.txt
Untracked: data/dosages_IITA_2021May13.rds
Untracked: data/dosages_IITA_filtered_2021May13.rds
Untracked: data/genmap_2021Aug02.rds
Untracked: data/genmap_2021May13.rds
Untracked: data/haps_IITA_2021May13.rds
Untracked: data/haps_IITA_filtered_2021May13.rds
Untracked: data/recombFreqMat_1minus2c_2021May13.rds
Untracked: output/Cassava_HMII_V3_Marning_imputation_6-18-21.samples
Untracked: output/IITA_CleanedTrialData_2021May10.rds
Untracked: output/IITA_ExptDesignsDetected_2021May10.rds
Untracked: output/IITA_blupsForModelTraining_twostage_asreml_2021May10.rds
Untracked: output/IITA_trials_NOT_identifiable.csv
Untracked: output/alphaAssignOutput_iita_pedigree.txt.dams
Untracked: output/alphaAssignOutput_iita_pedigree.txt.dams.full
Untracked: output/alphaAssignOutput_iita_pedigree.txt.pedigree
Untracked: output/alphaAssignOutput_iita_pedigree.txt.pedigree.top
Untracked: output/alphaAssignOutput_iita_pedigree.txt.sires
Untracked: output/alphaAssignOutput_iita_pedigree.txt.sires.full
Untracked: output/crossValPredsA.rds
Untracked: output/crossValPredsAD.rds
Untracked: output/cvAD_1rep_markerEffects.rds
Untracked: output/cvAD_1rep_meanPredAccuracy.rds
Untracked: output/cvAD_1rep_parentfolds.rds
Untracked: output/cvAD_1rep_predAccuracy.rds
Untracked: output/cvAD_1rep_predMeans.rds
Untracked: output/cvAD_1rep_predVars.rds
Untracked: output/cvAD_1rep_varPredAccuracy.rds
Untracked: output/cvAD_5rep5fold_markerEffects.rds
Untracked: output/cvAD_5rep5fold_meanPredAccuracy.rds
Untracked: output/cvAD_5rep5fold_parentfolds.rds
Untracked: output/cvAD_5rep5fold_predMeans.rds
Untracked: output/cvAD_5rep5fold_predVars.rds
Untracked: output/cvAD_5rep5fold_varPredAccuracy.rds
Untracked: output/cvDirDom_1Mb_50kb_pt5_predAccuracy.rds
Untracked: output/cvDirDom_1Mb_50kb_pt6_predAccuracy.rds
Untracked: output/cvDirDom_1Mb_50kb_pt8_predAccuracy.rds
Untracked: output/cvDirDom_1Mb_50kb_pt9_predAccuracy.rds
Untracked: output/cvDirDom_5rep5fold_markerEffects.rds
Untracked: output/cvDirDom_5rep5fold_meanPredAccuracy.rds
Untracked: output/cvDirDom_5rep5fold_parentfolds.rds
Untracked: output/cvDirDom_5rep5fold_predMeans.rds
Untracked: output/cvDirDom_5rep5fold_predVars.rds
Untracked: output/cvDirDom_5rep5fold_varPredAccuracy.rds
Untracked: output/cvMeanPredAccuracyA.rds
Untracked: output/cvMeanPredAccuracyAD.rds
Untracked: output/cvPredMeansA.rds
Untracked: output/cvPredMeansAD.rds
Untracked: output/cvVarPredAccuracyA.rds
Untracked: output/cvVarPredAccuracyAD.rds
Untracked: output/genomicMatePredictions_top121parents_ModelAD.csv
Untracked: output/genomicMatePredictions_top121parents_ModelAD.rds
Untracked: output/genomicMatePredictions_top121parents_ModelDirDom.csv
Untracked: output/genomicMatePredictions_top121parents_ModelDirDom.rds
Untracked: output/genomicPredictions_ModelAD.csv
Untracked: output/genomicPredictions_ModelAD.rds
Untracked: output/genomicPredictions_ModelDirDom.csv
Untracked: output/genomicPredictions_ModelDirDom.rds
Untracked: output/kinship_A_IITA_2021May13.rds
Untracked: output/kinship_D_IITA_2021May13.rds
Untracked: output/kinship_domGenotypic_IITA_2021July5.rds
Untracked: output/markEffsTest.rds
Untracked: output/markerEffects.rds
Untracked: output/markerEffectsA.rds
Untracked: output/markerEffectsAD.rds
Untracked: output/maxNOHAV_byStudy.csv
Untracked: output/obsCrossMeansAndVars.rds
Untracked: output/parentfolds.rds
Untracked: output/ped2check_genome.rds
Untracked: output/ped2genos.txt
Untracked: output/pednames2keep.txt
Untracked: output/pednames_Prune100_25_pt25.log
Untracked: output/pednames_Prune100_25_pt25.nosex
Untracked: output/pednames_Prune100_25_pt25.prune.in
Untracked: output/pednames_Prune100_25_pt25.prune.out
Untracked: output/potential_dams.txt
Untracked: output/potential_sires.txt
Untracked: output/predVarTest.rds
Untracked: output/samples2keep_IITA_2021May13.txt
Untracked: output/samples2keep_IITA_MAFpt01_prune50_25_pt98.log
Untracked: output/samples2keep_IITA_MAFpt01_prune50_25_pt98.nosex
Untracked: output/samples2keep_IITA_MAFpt01_prune50_25_pt98.prune.in
Untracked: output/samples2keep_IITA_MAFpt01_prune50_25_pt98.prune.out
Untracked: output/samples2keep_notInPHGdb.txt
Untracked: output/stdcvDirDom_1Mb_50kb_pt5_predAccuracy.rds
Untracked: output/stdcvDirDom_1Mb_50kb_pt6_predAccuracy.rds
Untracked: output/stdcvDirDom_1Mb_50kb_pt8_predAccuracy.rds
Untracked: output/stdcvDirDom_1Mb_50kb_pt9_predAccuracy.rds
Untracked: output/stdcv_AD_predAccuracy.rds
Untracked: output/stdcv_DirDom_predAccuracy.rds
Untracked: output/test_cvAD_markerEffects.rds
Untracked: output/test_cvAD_meanPredAccuracy.rds
Untracked: output/test_cvAD_parentfolds.rds
Untracked: output/test_cvAD_predAccuracy.rds
Untracked: output/test_cvAD_predMeans.rds
Untracked: output/test_cvAD_predVars.rds
Untracked: output/test_cvAD_varPredAccuracy.rds
Untracked: output/top50crosses.csv
Untracked: output/verified_ped.txt
Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes.
These are the previous versions of the repository in which changes were made to the R Markdown (analysis/05-CrossValidation.Rmd) and HTML (docs/05-CrossValidation.html) files. If you’ve configured a remote Git repository (see ?wflow_git_remote), click on the hyperlinks in the table below to view the files as they were in that past version.
| File | Version | Author | Date | Message |
|---|---|---|---|---|
| Rmd | f6f58f0 | wolfemd | 2021-08-04 | Tidy version (debugging stuff removed) of final analysis. |
| Rmd | 8362515 | wolfemd | 2021-08-04 | Completed debugging and successful run of cross-validation with reduced marker sets. Debugging work shown. |
| Rmd | a87d6d3 | wolfemd | 2021-08-03 | Optimize marker density for speed. Use LD pruning to make 4 subsets. Implement parent-wise and standard CV. All debugging work and full analyses shown. |
| html | ba8527c | wolfemd | 2021-08-02 | Build site. |
| Rmd | 511236f | wolfemd | 2021-08-02 | Simplify (remove debugging and prelim test sections) and organize for consistency. |
| Rmd | 2d5a542 | wolfemd | 2021-08-02 | Additional debugging of runCrossVal() completed with work shown. Full analysis also completed successfully. |
| Rmd | 230cd14 | wolfemd | 2021-07-30 | Develop upgraded runCrossVal() function to be integrated into code/gmsFunctions.R. Test is and then run full analysis. Debug work shown. |
| html | 769fe81 | wolfemd | 2021-07-29 | Build site. |
| Rmd | f7af63a | wolfemd | 2021-07-26 | Link to and add placeholders for extraction of a VCF from PHG DB and subsequent prepaation of inputs for subsequent analyses. |
| html | 934141c | wolfemd | 2021-07-14 | Build site. |
| html | cc1eb4b | wolfemd | 2021-07-14 | Build site. |
| Rmd | 772750a | wolfemd | 2021-07-14 | DirDom model and selection index calc fully integrated functions. |
| Rmd | 97db806 | wolfemd | 2021-07-11 | Work shown finding and fixing a bug where at least one getMarkEffs model failed. Problem was with use of plan(multicore) + OpenBLAS both using forking. Instead use plan(multisession). |
| Rmd | 2bc9644 | wolfemd | 2021-07-09 | Re-run cross-val with meanPredAccuracy SELIND handling fixed, but debug work not shown anymore. |
| Rmd | 889d98a | wolfemd | 2021-07-09 | test and fix bug in meanPredAccuracy() output when SIwts contain only subset of traits predicted. |
| Rmd | 4308b87 | wolfemd | 2021-07-08 | Full run 5-reps x 5-fold parent-wise cross-val both models DirDom and AD. |
| Rmd | 7888dee | wolfemd | 2021-07-08 | Work fully shown, testing and integrating DirDom model into crossval funcs. Now using R inside a singularity via rocker. Controlling OpenBLAS inside R session with RhpcBLASctl::blas_set_num_threads() and much more. |
| html | 5e45aac | wolfemd | 2021-06-18 | Build site. |
| Rmd | fa20501 | wolfemd | 2021-06-18 | Initial results are ready to publish and share with colleagues for |
| Rmd | 12cc368 | wolfemd | 2021-06-18 | runParentWiseCrossVal for 1 full rep, 5 folds. Found issue with CBSU R compilation but NOT with my code! |
| html | e66bdad | wolfemd | 2021-06-10 | Build site. |
| Rmd | a8452ba | wolfemd | 2021-06-10 | Initial build of the entire page upon completion of all |
| Rmd | 6a5ef32 | wolfemd | 2021-06-09 | meanPredAccuracy() now also included with function moved to “parentWiseCrossVal.R”. NOTE on previous commit: cross-validation functions are NOT in “predCrossVar.R”. |
| Rmd | 63067f7 | wolfemd | 2021-06-07 | Function varPredAccuracy() debugged / tested and moved to predCrossVar.R |
| Rmd | 66c0bde | wolfemd | 2021-06-07 | Remove old and unused code. STILL IN PROGRESS at the computeVarPredAccuracy step. |
| Rmd | 3c085ee | wolfemd | 2021-06-07 | Cross-validation code IN PROGRESS. Currently working on computeVarPredAccuracy. |
Also:
*.db file produced by Evan Long. Subsequently, prepare haplotype and dosage matrices, genetic map and recombination frequency matrix, for use in predictions.In the manuscript, the cross-validation is documented many pages and scripts, documented here.
For ongoing GS, I have a function runCrossVal() that manages all inputs and outputs easy to work with pre-computed accuracies.
Goal here is to make a function: runParentWiseCrossVal(), or at least make progress towards developing one.
However, for computational reasons, I imagine it might still be best to separate the task into a few functions.
My goal is to simplify and integrate into the pipeline used for NextGen Cassava. In the paper, used multi-trait Bayesian ridge-regression (MtBRR) to obtain marker effects, and also stored posterior matrices on disk to later compute posterior mean variances. This was computationally expensive and different from my standard univariate REML approach. I think MtBRR and PMV are probably the least biased way to go… but…
For the sake of testing a simple integration into the in-use pipeline, I want to try univariate REML to get the marker effects, which I’ll subsequently use for the cross-validation.
Revised the functions in package:predCrossVar to increase the computational efficiency. Not yet included into the actual R package but instead sourced from code/predCrossVar.R. Additional speed increases were achieved by extra testing to optimize balance of OMP_NUM_THREADS setting (multi-core BLAS) and parallel processing of the crosses-being-predicted. Improvements will benefit users predicting with REML / Bayesian-VPM, but probably worse for Bayesian-PMV.
Use a a singularity image from the rocker project, as recommended by Qi Sun to get an OpenBLAS-linked R environment that packages can easily be installed on.
This first chunk is one-time only and doesn’t take long. Saves a 650Mb *.sif file to server’s /workdir/
# copy the project data
cd /home/jj332_cas/marnin/;
cp -R implementGMSinCassava /home/$USER;
# the project directory can be in my networked folder for 2 reasons:
# 1) singularity will automatically recognize and be able to access it
# 2) My analyses not read/write intensive; don't break server rules/etiquette
# set up a working directory on the remote machine
mkdir /workdir/$USER
cd /workdir/$USER/;
# pull a singularity image and save in the file rocker.sif
# next time you use the rocker.sif file to start the container
singularity pull rocker.sif docker://rocker/tidyverse:latest;For analysis, operate each R session within a singularity Linux shell within a screen shell.
# 1) start a screen shell
screen; # or screen -r if re-attaching...
# 2) start the singularity Linux shell inside that
#singularity shell /workdir/$USER/rocker.sif;
singularity shell ~/rocker2.sif;
# Project directory, so R will use as working dir.
cd /home/mw489/implementGMSinCassava/;
# 3) Start R
RFully-tested runParentWiseCrossVal() and component functions are in the code/parentWiseCrossVal.R script.
Below, source it and use it for a full cross-validation run.
# install.packages(c("RhpcBLASctl","here","rsample","sommer","psych","future.callr","furrr","lme4","qs"))
# install.packages('future.callr')require(tidyverse); require(magrittr);
# 5 threads per Rsession for matrix math (openblas)
RhpcBLASctl::blas_set_num_threads(5)
# SOURCE CORE FUNCTIONS
source(here::here("code","parentWiseCrossVal.R"))
source(here::here("code","predCrossVar.R"))
# PEDIGREE
ped<-read.table(here::here("output","verified_ped.txt"),
header = T, stringsAsFactors = F) %>%
rename(GID=FullSampleName,
damID=DamID,
sireID=SireID) %>%
dplyr::select(GID,sireID,damID)
# Keep only families with _at least_ 2 offspring
ped %<>%
semi_join(ped %>% count(sireID,damID) %>% filter(n>1) %>% ungroup())
# BLUPs
blups<-readRDS(file=here::here("data","blups_forCrossVal.rds")) %>%
dplyr::select(-varcomp)
# GENOMIC RELATIONSHIP MATRICES (GRMS)
grms<-list(A=readRDS(file=here::here("output","kinship_A_IITA_2021May13.rds")),
D=readRDS(file=here::here("output",
"kinship_domGenotypic_IITA_2021July5.rds")))
## using A+domGenotypic (instead of domClassic used previously)
## will achieve appropriate dom effects for predicting family mean TGV
## but resulting add effects WILL NOT represent allele sub. effects and thus
## predictions won't equal GEBV, allele sub. effects will be post-computed
## as alpha = a + d(q-p)
# DOSAGE MATRIX
dosages<-readRDS(file=here::here("data",
"dosages_IITA_filtered_2021May13.rds"))
# RECOMBINATION FREQUENCY MATRIX
recombFreqMat<-readRDS(file=here::here("data",
"recombFreqMat_1minus2c_2021May13.rds"))
# HAPLOTYPE MATRIX
## keep only haplos for parents-in-the-pedigree
## those which will be used in prediction, saves memory
haploMat<-readRDS(file=here::here("data","haps_IITA_filtered_2021May13.rds"))
parents<-union(ped$sireID,ped$damID)
parenthaps<-sort(c(paste0(parents,"_HapA"),
paste0(parents,"_HapB")))
haploMat<-haploMat[parenthaps,colnames(recombFreqMat)]
# SELECTION INDEX WEIGHTS
## from IYR+IK
## note that not ALL predicted traits are on index
SIwts<-c(logFYLD=20,
HI=10,
DM=15,
MCMDS=-10,
logRTNO=12,
logDYLD=20,
logTOPYLD=15,
PLTHT=10) Server 1: modelType=“AD”
cbsulm29 - 104 cores, 512 GB RAM
grmsAD<-list(A=readRDS(file=here::here("output","kinship_A_IITA_2021May13.rds")),
D=readRDS(file=here::here("output",
"kinship_D_IITA_2021May13.rds")))
rm(grms)
cvAD_5rep5fold<-runParentWiseCrossVal(nrepeats=5,nfolds=5,seed=84,
modelType="AD",
ncores=20,
outName="output/cvAD_5rep5fold",
ped=ped,
blups=blups,
dosages=dosages,
haploMat=haploMat,
grms=grmsAD,
recombFreqMat = recombFreqMat,
selInd = TRUE, SIwts = SIwts)
saveRDS(cvAD_5rep5fold,here::here("output","cvAD_5rep5fold_predAccuracy.rds"))
# [1] "Marker-effects Computed. Took 1.81086 hrs"
# [1] "Done predicting fam vars. Took 43.11 mins for 198 crosses"
# [1] "Done predicting fam vars. Took 47.04 mins for 216 crosses"
# .....
# [1] "Accuracies predicted. Took 19.68694 hrs total.\n Goodbye!"
# [1] "Accuracies predicted. Took 19.73242 hrs total.Goodbye!"
# > saveRDS(cvAD_5rep5fold,here::here("output","cvAD_5rep5fold_predAccuracy.rds"))Server 2: modelType=“DirDom”
cbsulm17 - 112 cores, 512 GB RAM
cvDirDom_5rep5fold<-runParentWiseCrossVal(nrepeats=5,nfolds=5,seed=84,
modelType="DirDom",
ncores=20,nBLASthreads=5,
outName="output/cvDirDom_5rep5fold",
ped=ped,
blups=blups,
dosages=dosages,
haploMat=haploMat,
grms=grms,
recombFreqMat = recombFreqMat,
selInd = TRUE, SIwts = SIwts)
saveRDS(cvDirDom_5rep5fold,here::here("output","cvDirDom_5rep5fold_predAccuracy.rds"))
# [1] "Marker-effects Computed. Took 2.3851 hrs"
# [1] "Predicting cross variances and covariances"
# Joining, by = c("Repeat", "Fold")
# [1] "Done predicting fam vars. Took 59.08 mins for 198 crosses"
# [1] "Done predicting fam vars. Took 18.63 mins for 198 crosses"
# [1] "Done predicting fam vars. Took 64.82 mins for 216 crosses"
# [1] "Done predicting fam vars. Took 20.41 mins for 216 crosses"
# [1] "Done predicting fam vars. Took 46.42 mins for 156 crosses"
# [1] "Done predicting fam vars. Took 14.94 mins for 156 crosses"
# [1] "Done predicting fam vars. Took 63.45 mins for 210 crosses"
# [1] "Done predicting fam vars. Took 19.8 mins for 210 crosses"
# [1] "Done predicting fam vars. Took 50.62 mins for 171 crosses"
# [1] "Done predicting fam vars. Took 16.26 mins for 171 crosses"
# [1] "Done predicting fam vars. Took 49.87 mins for 163 crosses"
# [1] "Done predicting fam vars. Took 16.2 mins for 163 crosses"
# [1] "Done predicting fam vars. Took 73.37 mins for 253 crosses"
# [1] "Done predicting fam vars. Took 23.59 mins for 253 crosses"
# [1] "Done predicting fam vars. Took 56.32 mins for 190 crosses"
# [1] "Done predicting fam vars. Took 18.44 mins for 190 crosses"
# [1] "Done predicting fam vars. Took 47.33 mins for 161 crosses"
# [1] "Done predicting fam vars. Took 15.79 mins for 161 crosses"
# [1] "Done predicting fam vars. Took 59.18 mins for 189 crosses"
# [1] "Done predicting fam vars. Took 18.67 mins for 189 crosses"
# [1] "Done predicting fam vars. Took 64.72 mins for 205 crosses"
# [1] "Done predicting fam vars. Took 21.17 mins for 205 crosses"
# [1] "Done predicting fam vars. Took 63.97 mins for 213 crosses"
# [1] "Done predicting fam vars. Took 20.04 mins for 213 crosses"
# [1] "Done predicting fam vars. Took 53.03 mins for 180 crosses"
# [1] "Done predicting fam vars. Took 17.28 mins for 180 crosses"
# [1] "Done predicting fam vars. Took 58.67 mins for 199 crosses"
# [1] "Done predicting fam vars. Took 19.03 mins for 199 crosses"
# ....
# estimate 20 more hours, complete on July 12 very early AM?
# [1] "Accuracies predicted. Took 34.37369 hrs total.Goodbye!"
# Warning message:
# In for (ii in 1L:length(res)) { : closing unused connection 3 (localhost)
# > saveRDS(cvDirDom_5rep5fold,here::here("output","cvDirDom_5rep5fold_predAccuracy.rds"))The new “parent-wise” cross-validation scheme assesses the accuracy of predicting the means and variances of crosses.
The “standard” cross-validation, used in all previous genomic selection analyses (e.g. IITA_2020GS CV, NRCRI_2021GS CV, TARI_2020GS CV), assesses the accuracy predicting the individual performance (breeding value or TGV).
[NEW]: Below, I upgrade the runCrossVal() function used previously for “standard” cross-validation. Include selection index (via selInd= and SIwts= arguments) and a modelType="DirDom" option.
# 1) start a screen shell
screen; # or screen -r if re-attaching...
# 2) start the singularity Linux shell inside that
#singularity shell /workdir/$USER/rocker.sif;
singularity shell ~/rocker2.sif;
# Project directory, so R will use as working dir.
cd /home/mw489/implementGMSinCassava/;
# 3) Start R
Rrequire(tidyverse); require(magrittr);
# SOURCE CORE FUNCTIONS
source(here::here("code","gmsFunctions.R"))
source(here::here("code","predCrossVar.R"))
# BLUPs
blups<-readRDS(file=here::here("data","blups_forCrossVal.rds")) %>%
dplyr::select(-varcomp) %>%
rename(TrainingData=blups) # for compatibility with downstream functions
# SELECTION INDEX WEIGHTS
## from IYR+IK
## note that not ALL predicted traits are on index
SIwts<-c(logFYLD=20,
HI=10,
DM=15,
MCMDS=-10,
logRTNO=12,
logDYLD=20,
logTOPYLD=15,
PLTHT=10) cbsulm15 - 112 cores, 512 GB RAM
# GENOMIC RELATIONSHIP MATRICES (GRMS)
grms_ad<-list(A=readRDS(file=here::here("output","kinship_A_IITA_2021May13.rds")),
D=readRDS(file=here::here("output","kinship_D_IITA_2021May13.rds")))
## achieves the classic TGV=BV+TGV partition
stdcv_ad<-runCrossVal(blups=blups,
modelType="AD",
selInd=TRUE,SIwts=SIwts,
grms=grms_ad,dosages=NULL,
nrepeats=5,nfolds=5,
ncores=20,nBLASthreads=5,
gid="GID",seed=42)
saveRDS(stdcv_ad,here::here("output","stdcv_AD_predAccuracy.rds"))
stdcv_ad %>%
select(-splits) %>%
unnest(accuracyEstOut) %>%
select(-predVSobs)cbsulm17 - 112 cores, 512 GB RAM
# GENOMIC RELATIONSHIP MATRICES (GRMS)
grms_dirdom<-list(A=readRDS(file=here::here("output","kinship_A_IITA_2021May13.rds")),
D=readRDS(file=here::here("output",
"kinship_domGenotypic_IITA_2021July5.rds")))
## using A+domGenotypic (instead of domClassic used previously)
## will achieve appropriate dom effects for predicting family mean TGV
## but resulting add effects WILL NOT represent allele sub. effects and thus
## predictions won't equal GEBV, allele sub. effects will be post-computed
## as alpha = a + d(q-p)
# DOSAGE MATRIX
dosages<-readRDS(file=here::here("data",
"dosages_IITA_filtered_2021May13.rds"))
stdcv_dirdom<-runCrossVal(blups=blups,
modelType="DirDom",
selInd=TRUE,SIwts=SIwts,
grms=grms_dirdom,dosages=dosages,
nrepeats=5,nfolds=5,
ncores=20,nBLASthreads=5,
gid="GID",seed=42)
saveRDS(stdcv_dirdom,here::here("output","stdcv_DirDom_predAccuracy.rds"))# 1) start a screen shell
screen; # or screen -r if re-attaching...
# 2) start the singularity Linux shell inside that
#singularity shell /workdir/$USER/rocker.sif;
singularity shell ~/rocker2.sif;
# Project directory, so R will use as working dir.
cd /home/mw489/implementGMSinCassava/;
# 3) Start R
RCross-variance prediction is slow, but significant speed gains can be made by using fewer markers for the predictions. Examine the speed benefit vs. accuracy cost trade-off.
Two ways to accomplish reduced marker number: (1) random sample, (2) LD-pruning (as in the preprocessing step here).
Try both below.
Use cross-validation to compare accuracy achieved.
DirDom model-only at this point.
Use more stringent versions of the plink1.9 --indep-pairwise used previously (here)
cd ~/implementGMSinCassava/
export PATH=/programs/plink-1.9-x86_64-beta3.30:$PATH;
# plink --bfile data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719 \
# --keep output/samples2keep_IITA_2021May13.txt \
# --indep-pairwise 250 'kb' 125 0.9 \
# --out output/samples2keep_IITA_MAFpt01_prune250kb_125_pt9;
# 24986 of 68068 variants removed. (43082 left, more than original filter)
plink --bfile data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719 \
--keep output/samples2keep_IITA_2021May13.txt \
--indep-pairwise 1000 'kb' 50 0.9 \
--out output/samples2keep_IITA_MAFpt01_prune1Mb_50kb_pt9;
# Pruned 6587 variants from chromosome 1, leaving 1720.
# Pruned 2524 variants from chromosome 2, leaving 1457.
# Pruned 2904 variants from chromosome 3, leaving 1350.
# Pruned 3110 variants from chromosome 4, leaving 1006.
# Pruned 2918 variants from chromosome 5, leaving 1222.
# Pruned 2510 variants from chromosome 6, leaving 1142.
# Pruned 1288 variants from chromosome 7, leaving 800.
# Pruned 2454 variants from chromosome 8, leaving 1311.
# Pruned 2557 variants from chromosome 9, leaving 971.
# Pruned 1889 variants from chromosome 10, leaving 1123.
# Pruned 2225 variants from chromosome 11, leaving 1175.
# Pruned 1982 variants from chromosome 12, leaving 990.
# Pruned 2059 variants from chromosome 13, leaving 879.
# Pruned 4004 variants from chromosome 14, leaving 1540.
# Pruned 2877 variants from chromosome 15, leaving 1097.
# Pruned 2059 variants from chromosome 16, leaving 835.
# Pruned 1996 variants from chromosome 17, leaving 976.
# Pruned 2226 variants from chromosome 18, leaving 1051.
# Pruning complete. 48169 of 68814 variants removed. (20645 left)
plink --bfile data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719 \
--keep output/samples2keep_IITA_2021May13.txt \
--indep-pairwise 1000 'kb' 50 0.8 \
--out output/samples2keep_IITA_MAFpt01_prune1Mb_50kb_pt8;
# Pruned 7131 variants from chromosome 1, leaving 1176.
# Pruned 2888 variants from chromosome 2, leaving 1093.
# Pruned 3287 variants from chromosome 3, leaving 967.
# Pruned 3358 variants from chromosome 4, leaving 758.
# Pruned 3229 variants from chromosome 5, leaving 911.
# Pruned 2732 variants from chromosome 6, leaving 920.
# Pruned 1438 variants from chromosome 7, leaving 650.
# Pruned 2738 variants from chromosome 8, leaving 1027.
# Pruned 2826 variants from chromosome 9, leaving 702.
# Pruned 2190 variants from chromosome 10, leaving 822.
# Pruned 2506 variants from chromosome 11, leaving 894.
# Pruned 2236 variants from chromosome 12, leaving 736.
# Pruned 2268 variants from chromosome 13, leaving 670.
# Pruned 4448 variants from chromosome 14, leaving 1096.
# Pruned 3188 variants from chromosome 15, leaving 786.
# Pruned 2256 variants from chromosome 16, leaving 638.
# Pruned 2247 variants from chromosome 17, leaving 725.
# Pruned 2506 variants from chromosome 18, leaving 771.
# Pruning complete. 53472 of 68814 variants removed. (15342 left)
plink --bfile data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719 \
--keep output/samples2keep_IITA_2021May13.txt \
--indep-pairwise 1000 'kb' 50 0.7 \
--out output/samples2keep_IITA_MAFpt01_prune1Mb_50kb_pt7;
# Pruned 7413 variants from chromosome 1, leaving 894.
# Pruned 3118 variants from chromosome 2, leaving 863.
# Pruned 3504 variants from chromosome 3, leaving 750.
# Pruned 3484 variants from chromosome 4, leaving 632.
# Pruned 3426 variants from chromosome 5, leaving 714.
# Pruned 2913 variants from chromosome 6, leaving 739.
# Pruned 1555 variants from chromosome 7, leaving 533.
# Pruned 2932 variants from chromosome 8, leaving 833.
# Pruned 2972 variants from chromosome 9, leaving 556.
# Pruned 2372 variants from chromosome 10, leaving 640.
# Pruned 2709 variants from chromosome 11, leaving 691.
# Pruned 2381 variants from chromosome 12, leaving 591.
# Pruned 2411 variants from chromosome 13, leaving 527.
# Pruned 4736 variants from chromosome 14, leaving 808.
# Pruned 3370 variants from chromosome 15, leaving 604.
# Pruned 2372 variants from chromosome 16, leaving 522.
# Pruned 2396 variants from chromosome 17, leaving 576.
# Pruned 2663 variants from chromosome 18, leaving 614.
# Pruning complete. 56727 of 68814 variants removed. (12087 left)
plink --bfile data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719 \
--keep output/samples2keep_IITA_2021May13.txt \
--indep-pairwise 1000 'kb' 50 0.6 \
--out output/samples2keep_IITA_MAFpt01_prune1Mb_50kb_pt6;
# Pruned 7619 variants from chromosome 1, leaving 688.
# Pruned 3354 variants from chromosome 2, leaving 627.
# Pruned 3666 variants from chromosome 3, leaving 588.
# Pruned 3596 variants from chromosome 4, leaving 520.
# Pruned 3597 variants from chromosome 5, leaving 543.
# Pruned 3060 variants from chromosome 6, leaving 592.
# Pruned 1670 variants from chromosome 7, leaving 418.
# Pruned 3094 variants from chromosome 8, leaving 671.
# Pruned 3073 variants from chromosome 9, leaving 455.
# Pruned 2511 variants from chromosome 10, leaving 501.
# Pruned 2882 variants from chromosome 11, leaving 518.
# Pruned 2493 variants from chromosome 12, leaving 479.
# Pruned 2505 variants from chromosome 13, leaving 433.
# Pruned 4929 variants from chromosome 14, leaving 615.
# Pruned 3528 variants from chromosome 15, leaving 446.
# Pruned 2455 variants from chromosome 16, leaving 439.
# Pruned 2508 variants from chromosome 17, leaving 464.
# Pruned 2773 variants from chromosome 18, leaving 504.
# Pruning complete. 59313 of 68814 variants removed. (9501 left)
plink --bfile data/AllChrom_RefPanelAndGSprogeny_ReadyForGP_72719 \
--keep output/samples2keep_IITA_2021May13.txt \
--indep-pairwise 1000 'kb' 50 0.5 \
--out output/samples2keep_IITA_MAFpt01_prune1Mb_50kb_pt5;
# Pruned 7791 variants from chromosome 1, leaving 516.
# Pruned 3497 variants from chromosome 2, leaving 484.
# Pruned 3800 variants from chromosome 3, leaving 454.
# Pruned 3688 variants from chromosome 4, leaving 428.
# Pruned 3722 variants from chromosome 5, leaving 418.
# Pruned 3205 variants from chromosome 6, leaving 447.
# Pruned 1747 variants from chromosome 7, leaving 341.
# Pruned 3228 variants from chromosome 8, leaving 537.
# Pruned 3161 variants from chromosome 9, leaving 367.
# Pruned 2622 variants from chromosome 10, leaving 390.
# Pruned 2992 variants from chromosome 11, leaving 408.
# Pruned 2589 variants from chromosome 12, leaving 383.
# Pruned 2591 variants from chromosome 13, leaving 347.
# Pruned 5094 variants from chromosome 14, leaving 450.
# Pruned 3653 variants from chromosome 15, leaving 321.
# Pruned 2549 variants from chromosome 16, leaving 345.
# Pruned 2596 variants from chromosome 17, leaving 376.
# Pruned 2877 variants from chromosome 18, leaving 400.
# Pruning complete. 61402 of 68814 variants removed. (7412)Settled on 4 subsets based on LD-pruning plink --indep-pairwise looking at 1 Mb windows in steps of 50 Kb pruning using the \(r^2\) thresholds 0.9 (~20K), 0.8 (~15K), 0.6 (~10K), 0.5 (~7K). Skipping 0.7 because 12K just not diff enough from 10 or 15K for me.
Used plink-output list below to prune downstream input files.
Subsetting the unfiltered version of the haps and dosages (e.g. dosages_IITA_2021May13.rds) can be done on read-in rather than creating new, smaller files on disk.
grms (kinship matrices) will need to be remade and pre-stored on disk.recombFreqMat needs to be created and stored on disk based on that. After that the recombFreqMat can also be subset on loading.library(tidyverse); library(magrittr);
dosages<-readRDS(file=here::here("data","dosages_IITA_2021May13.rds"))
source(here::here("code","gmsFunctions.R"))
snpsets<-tibble(Filter=c(5,6,8,9)) %>%
mutate(snps2keep=map(Filter,~read.table(here::here("output",
paste0("samples2keep_IITA_MAFpt01_prune1Mb_50kb_pt",.,".prune.in")),
header = F, stringsAsFactors = F)))
snpsets %<>%
mutate(snps2keep=map(snps2keep,function(snps2keep,...){
tokeep<-tibble(FULL_SNP_ID=colnames(dosages)) %>%
separate(FULL_SNP_ID,c("Chr","Pos","Ref","Alt"),remove = F) %>%
mutate(SNP_ID=paste0("S",Chr,"_",Pos)) %>%
filter(SNP_ID %in% snps2keep$V1)
return(tokeep)}),
Filter=paste0("1Mb_50kb_pt",Filter))
RhpcBLASctl::blas_set_num_threads(56)
snpsets %<>%
mutate(Amat=map(snps2keep,~kinship(dosages[,.$FULL_SNP_ID],type="add")),
Dmat=map(snps2keep,~kinship(dosages[,.$FULL_SNP_ID],type="domGenotypic")))
library(qs)
qsave(snpsets,file=here::here("output","kinships_LDpruningSeries_2021Aug02.qs"),nthreads = 20)Next interpolate a genetic map for all 68K markers for max utility. As mentioned above, need it to make a recombFreqMat.
genmap<-read.table(here::here("data/CassavaGeneticMap",
"cassava_cM_pred.v6.allchr.txt"),
header = F, stringsAsFactors = F,sep=';') %>%
rename(SNP_ID=V1,Pos=V2,cM=V3) %>%
as_tibble
snps_genmap<-tibble(DoseSNP_ID=colnames(dosages)) %>%
separate(DoseSNP_ID,c("Chr","Pos","Ref","Alt"),remove = F) %>%
mutate(SNP_ID=paste0("S",Chr,"_",Pos)) %>%
left_join(genmap %>% mutate(across(everything(),as.character)))interpolate_genmap<-function(data){
# for each chromosome map
# find and _decrements_ in the genetic map distance
# fix them to the cumulative max to force map to be only increasing
# fit a spline for each chromosome
# Use it to predict values for positions not previously on the map
# fix them AGAIN (in case) to the cumulative max, forcing map to only increase
data_forspline<-data %>%
filter(!is.na(cM)) %>%
mutate(cumMax=cummax(cM),
cumIncrement=cM-cumMax) %>%
filter(cumIncrement>=0) %>%
select(-cumMax,-cumIncrement)
spline<-data_forspline %$% smooth.spline(x=Pos,y=cM,spar = 0.75)
splinemap<-predict(spline,x = data$Pos) %>%
as_tibble(.) %>%
rename(Pos=x,cM=y) %>%
mutate(cumMax=cummax(cM),
cumIncrement=cM-cumMax) %>%
mutate(cM=cumMax) %>%
select(-cumMax,-cumIncrement)
return(splinemap)
}splined_snps_genmap<-snps_genmap %>%
select(-cM) %>%
mutate(Pos=as.numeric(Pos)) %>%
left_join(snps_genmap %>%
mutate(across(c(Pos,cM),as.numeric)) %>%
arrange(Chr,Pos) %>%
nest(data=-Chr) %>%
mutate(data=map(data,interpolate_genmap)) %>%
unnest(data)) %>%
distinctall(splined_snps_genmap$DoseSNP_ID == colnames(dosages))
# [1] TRUE
saveRDS(splined_snps_genmap,file=here::here("data","genmap_2021Aug02.rds"))readRDS(here::here("data","genmap_2021Aug02.rds")) %>%
mutate(Map="Spline") %>%
ggplot(.,aes(x=Pos/1000/1000,y=cM,color=Map),alpha=0.5,size=0.75) +
geom_point() +
theme_bw() + facet_wrap(~as.integer(Chr), scales='free_x')Construct a matrix of recombination frequencies at loci for all study loci. Pre-compute 1-2c to save time predicting cross variance.
rm(list=ls()); gc()
library(tidyverse); library(magrittr); library(qs)
RhpcBLASctl::blas_set_num_threads(106)
source(here::here("code","predCrossVar.R"))
genmap<-readRDS(file=here::here("data","genmap_2021Aug02.rds"))
m<-genmap$cM;
names(m)<-genmap$DoseSNP_ID
recombFreqMat<-1-(2*genmap2recombfreq(m,nChr = 18))
qsave(recombFreqMat,file=here::here("data","recombFreqMat_1minus2c_2021Aug02.qs"),nthreads = 56)require(tidyverse); require(magrittr); library(qs)
# SOURCE CORE FUNCTIONS
source(here::here("code","parentWiseCrossVal.R"))
source(here::here("code","predCrossVar.R"))
# PEDIGREE
ped<-read.table(here::here("output","verified_ped.txt"),
header = T, stringsAsFactors = F) %>%
rename(GID=FullSampleName,
damID=DamID,
sireID=SireID) %>%
dplyr::select(GID,sireID,damID)
# Keep only families with _at least_ 2 offspring
ped %<>%
semi_join(ped %>% count(sireID,damID) %>% filter(n>1) %>% ungroup())
# BLUPs
blups<-readRDS(file=here::here("data","blups_forCrossVal.rds")) %>%
dplyr::select(-varcomp)
# LD-PRUNNED SNP SETS + GRMS
snpsets<-qread(here::here("output","kinships_LDpruningSeries_2021Aug02.qs"),
nthreads = 56)
# DOSAGE MATRIX (UNFILTERED)
dosages<-readRDS(file=here::here("data","dosages_IITA_2021May13.rds"))
# RECOMBINATION FREQUENCY MATRIX (UNFILTERED)
recombFreqMat<-qread(file=here::here("data",
"recombFreqMat_1minus2c_2021Aug02.qs"))
# ,
# nthreads = 56)
# HAPLOTYPE MATRIX (UNFILTERED)
## keep only haplos for parents-in-the-pedigree
## those which will be used in prediction, saves memory
haploMat<-readRDS(file=here::here("data","haps_IITA_2021May13.rds"))
parents<-union(ped$sireID,ped$damID)
parenthaps<-sort(c(paste0(parents,"_HapA"),
paste0(parents,"_HapB")))
haploMat<-haploMat[parenthaps,colnames(recombFreqMat)]
# SELECTION INDEX WEIGHTS
## from IYR+IK
## note that not ALL predicted traits are on index
SIwts<-c(logFYLD=20,
HI=10,
DM=15,
MCMDS=-10,
logRTNO=12,
logDYLD=20,
logTOPYLD=15,
PLTHT=10) Note: using same seed and other params as original run with 34K SNPs, which took ~34hrs.
# cbsulm15 - 112 cores, 512 GB RAM - 2021 Aug 3 - 4:20pm
filterLevel<-"1Mb_50kb_pt5"
# [1] "Accuracies predicted. Took 4.42406 hrs total.Goodbye!"
# [1] "Time elapsed: 265.469 mins"
# cbsulm17 - 112 cores, 512 GB RAM - 2021 Aug 3 - " "
filterLevel<-"1Mb_50kb_pt6"
# [1] "Accuracies predicted. Took 5.28384 hrs total.Goodbye!"
# [1] "Time elapsed: 317.053 mins"
# cbsulm29 - 104 cores, 512 GB RAM - 2021 Aug 3 - " "
filterLevel<-"1Mb_50kb_pt8"
# [1] "Accuracies predicted. Took 7.14035 hrs total.Goodbye!"
# [1] "Time elapsed: 428.437 mins"
# cbsulm31 - 112 cores, 512 GB RAM - 2021 Aug 3 - " "
filterLevel<-"1Mb_50kb_pt9"
# [1] "Accuracies predicted. Took 10.65117 hrs total.Goodbye!"
# [1] "Time elapsed: 639.087 mins"snps2keep<-snpsets %>%
filter(Filter==filterLevel) %>%
select(snps2keep) %>%
unnest(snps2keep)
grms<-list(A=snpsets %>% filter(Filter==filterLevel) %$% Amat[[1]],
D=snpsets %>% filter(Filter==filterLevel) %$% Dmat[[1]])
dosages<-dosages[,snps2keep$FULL_SNP_ID]
haploMat<-haploMat[,snps2keep$FULL_SNP_ID]
recombFreqMat<-recombFreqMat[snps2keep$FULL_SNP_ID,snps2keep$FULL_SNP_ID]
rm(snpsets); gc()
starttime<-proc.time()[3]
cvDirDom<-runParentWiseCrossVal(nrepeats=5,nfolds=5,seed=84,
modelType="DirDom",
ncores=20,nBLASthreads=5,
outName=NULL,
ped=ped,
blups=blups,
dosages=dosages,
haploMat=haploMat,
grms=grms,
recombFreqMat = recombFreqMat,
selInd = TRUE, SIwts = SIwts)
saveRDS(cvDirDom,
file = here::here("output",
paste0("cvDirDom_",filterLevel,"_predAccuracy.rds")))
endtime<-proc.time()[3]; print(paste0("Time elapsed: ",round((endtime-starttime)/60,3)," mins"))require(tidyverse); require(magrittr); library(qs)
# SOURCE CORE FUNCTIONS
source(here::here("code","gmsFunctions.R"))
source(here::here("code","predCrossVar.R"))
# BLUPs
blups<-readRDS(file=here::here("data","blups_forCrossVal.rds")) %>%
dplyr::select(-varcomp) %>%
rename(TrainingData=blups) # for compatibility with downstream functions
# SELECTION INDEX WEIGHTS
## from IYR+IK
## note that not ALL predicted traits are on index
SIwts<-c(logFYLD=20,
HI=10,
DM=15,
MCMDS=-10,
logRTNO=12,
logDYLD=20,
logTOPYLD=15,
PLTHT=10)
# LD-PRUNNED SNP SETS + GRMS
snpsets<-qread(here::here("output","kinships_LDpruningSeries_2021Aug02.qs"),
nthreads = 56)
# DOSAGE MATRIX (UNFILTERED)
dosages<-readRDS(file=here::here("data","dosages_IITA_2021May13.rds"))# running...
# cbsulm18 - 88 cores, 512 GB RAM - 2021 Aug 3 - 1pm
filterLevel<-"1Mb_50kb_pt5" # [1] "Time elapsed: 131.09 mins"
# cbsulm18 - 88 cores, 512 GB RAM - 2021 Aug 3 - 4pm
filterLevel<-"1Mb_50kb_pt6" # [1] "Time elapsed: 131.54 mins"
# cbsulm18 - 88 cores, 512 GB RAM - 2021 Aug 3 - 7:15pm
filterLevel<-"1Mb_50kb_pt8" # [1] "Time elapsed: 136.943 mins"
# cbsulm18 - 88 cores, 512 GB RAM - 2021 Aug 3 - 10:20pm
filterLevel<-"1Mb_50kb_pt9"snps2keep<-snpsets %>%
filter(Filter==filterLevel) %>%
select(snps2keep) %>%
unnest(snps2keep)
grms<-list(A=snpsets %>% filter(Filter==filterLevel) %$% Amat[[1]],
D=snpsets %>% filter(Filter==filterLevel) %$% Dmat[[1]])
dosages<-dosages[,snps2keep$FULL_SNP_ID]
rm(snpsets); gc()
starttime<-proc.time()[3]
stdcv<-runCrossVal(blups=blups,
modelType="DirDom",
selInd=TRUE,SIwts=SIwts,
grms=grms,dosages=dosages,
nrepeats=5,nfolds=5,
ncores=17,nBLASthreads=5,
gid="GID",seed=42)
saveRDS(stdcv,
here::here("output",
paste0("stdcvDirDom_",filterLevel,"_predAccuracy.rds")))
endtime<-proc.time()[3]; print(paste0("Time elapsed: ",round((endtime-starttime)/60,3)," mins"))See Results: Home for plots and summary tables.
sessionInfo()