06-GenomicPredictions.Rmd

---
title: "Genomic predictions"
author: "Marnin Wolfe"
date: "2021-July-08"
output: 
  workflowr::wflow_html:
    toc: true
editor_options:
  chunk_output_type: inline
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = F, 
                      eval = FALSE, # <- NOTE THAT EVAL SET TO FALSE!
                      tidy='styler', tidy.opts=list(strict=FALSE,width.cutoff=100), highlight=TRUE)
```

# Previous step

5.  [Parent-wise cross-validation](05-CrossValidation.html): Compute parent-wise cross-validation folds using the validated pedigree. Fit models to get marker effects and make subsequent predictions of cross means and (co)variances.

# Current steps

1.  **Genomic prediction of clone GEBV/GETGV.** Fit GBLUP model, using genotypic add-dom partition. NEW: modelType="DirDom", include genome-wide inbreeding effect in GEBV/GETGV predictions *after* backsolving SNP effects. For all models, extract GBLUPs and backsolve SNP effects for use in cross usefulness predictions (mean+variance predictions). **ALSO NEW**: selection index predictions.

2.  **Genomic prediction of cross** $UC_{parent}$ and $UC_{variety}$. Rank potential parents on SI. Predict all possible crosses of some portion of best parents.

# Genomic prediction of clone GEBV/GETGV

```{bash Renv, eval=F}
# 1) start a screen shell 
screen; # or screen -r if re-attaching...
# 2) start the singularity Linux shell inside that
singularity shell ~/rocker2.sif; 
# Project directory, so R will use as working dir.
cd /home/mw489/IITA_2021GS/;
# 3) Start R
R
```

Load input data

```{r inputs for GP}
# NEEDED LIBRARIES
require(tidyverse); require(magrittr); library(qs)
library(genomicMateSelectR)

# BLUPs
blups<-readRDS(file=here::here("data","blups_forGP.rds")) %>% 
  dplyr::select(-varcomp) %>% 
  rename(TrainingData=blups) # for compatibility with runCrossVal() function 

# DOSAGE MATRIX (UNFILTERED)
dosages<-readRDS(file=here::here("data","dosages_IITA_2021Aug09.rds"))

# SNP SETS TO ANALYZE
snpsets<-readRDS(file = here::here("data","snpsets.rds"))

# 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) 
```

Run the "DirDom" modelTypes built into `runGenomicPredictions()`. Output will contain both GBLUPs for selection of clones _and_ SNP effects to use as input for prediction of cross usefulness and subsequent mate selection. 

Get effects for: **full_set** (~31K SNPs), **medium_set** (~13K, LD-pruned) and **reduced_set** (~9K SNPs, LD-pruned).

```{r GP analysis chunks}
# cbsulm30 - 112 cores, 512 GB RAM - 2021 Aug 10 - 8:40am 
snpSet<-"full_set"
grms<-list(A=readRDS(here::here("output","kinship_A_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_IITA_2021Aug09.rds")))
# [1] "Time elapsed: 24.066 mins"

# cbsulm30 - 112 cores, 512 GB RAM - 2021 Aug 10 - 8:40am
snpSet<-"reduced_set"
grms<-list(A=readRDS(here::here("output","kinship_A_ReducedSNPset_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_ReducedSNPset_IITA_2021Aug09.rds")))
# [1] "Time elapsed: 22.431 mins"

# cbsulm20 - 88 cores, 512 GB RAM - 2021 Aug 10 - 2:50pm
snpSet<-"medium_set"
grms<-list(A=readRDS(here::here("output","kinship_A_MediumSNPset_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_MediumSNPset_IITA_2021Aug09.rds")))
# [1] "Time elapsed: 13.299 mins"

```
```{r run GP for each snpset}
snps2keep<-snpsets %>% 
  filter(Set==snpSet) %>% 
  select(snps2keep) %>% 
  unnest(snps2keep)

dosages<-dosages[,snps2keep$FULL_SNP_ID]
rm(snpsets); gc()

starttime<-proc.time()[3]
gpreds<-runGenomicPredictions(modelType="DirDom",selInd=TRUE, SIwts=SIwts,
                              getMarkEffs=TRUE,
                              returnPEV=FALSE,
                              blups=blups,grms=grms,dosages=dosages,
                              ncores=11,nBLASthreads=5)
saveRDS(gpreds,
        file = here::here("output",
                          paste0("genomicPredictions_",snpSet,"_2021Aug09.rds")))
endtime<-proc.time()[3]; print(paste0("Time elapsed: ",
                                      round((endtime-starttime)/60,3)," mins"))

```

# Genomic prediction of cross usefulness

1. Use GBLUPs on SELIND to choose top fraction of the clones. **Cross-reference with 'accessions_infield' list**.
2. For those selected parents, predict the SELIND usefulness for all pairwise matings.

```{bash Renv for crossPreds, eval=F}
# 1) start a screen shell 
screen; # or screen -r if re-attaching...
# 2) start the singularity Linux shell inside that
singularity shell ~/rocker2.sif; 
# Project directory, so R will use as working dir.
cd /home/mw489/IITA_2021GS/;
# 3) Start R
R
```


```{r crossPred input}
# NEEDED LIBRARIES
require(tidyverse); require(magrittr); library(qs)
library(genomicMateSelectR)

# BLUPs
blups<-readRDS(file=here::here("data","blups_forGP.rds")) %>% 
  dplyr::select(-varcomp)

# DOSAGE MATRIX (UNFILTERED)
dosages<-readRDS(file=here::here("data","dosages_IITA_2021Aug09.rds"))

# RECOMBINATION FREQUENCY MATRIX (UNFILTERED)
recombFreqMat<-qread(file=here::here("data",
                                     "recombFreqMat_1minus2c_2021Aug02.qs"))

# 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_2021Aug09.rds"))
parents<-union(ped$sireID,ped$damID) 
parenthaps<-sort(c(paste0(parents,"_HapA"),
                   paste0(parents,"_HapB")))
haploMat<-haploMat[parenthaps,]

# SNP SETS TO ANALYZE
snpsets<-readRDS(file = here::here("data","snpsets.rds"))

# 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) 

# LIST OF ACCESSIONS LIKELY IN THE FIELD
accessions_infield<-readRDS(here::here("data",
                                       "accessions_possibly_infield_2021Aug10.rds"))

```

```{r cor of GBLUPs among snpsets, eval=F}
# gpreds_full<-readRDS(file = here::here("output","genomicPredictions_full_set_2021Aug09.rds"))
# gpreds_medium<-readRDS(file = here::here("output","genomicPredictions_medium_set_2021Aug09.rds"))
# gpreds_reduced<-readRDS(file = here::here("output","genomicPredictions_reduced_set_2021Aug09.rds"))
### Quick check that GBLUPs from full and medium_set are strongly correlated
### This will be an additional assurance that similar cross variances
### will be predicted by the reduced SNP model

### Cor between SELIND GEBV and GETGV between full_set and reduced_set of SNPs?
# left_join(gpreds_full$gblups[[1]] %>% select(GID,predOf,SELIND) %>% rename(SELIND_full=SELIND),
#           gpreds_reduced$gblups[[1]] %>% select(GID,predOf,SELIND) %>% rename(SELIND_reduced=SELIND)) %>%
#   group_by(predOf) %>% 
#   summarize(SELIND_corModels=cor(SELIND_full,SELIND_reduced))
# predOf  SELIND_corModels
# GEBV	0.0363334			
# GETGV	0.7809016	
## TERRIBLE!! :(

### Cor between SELIND GEBV and GETGV between full_set and medium_set of SNPs?
# left_join(gpreds_full$gblups[[1]] %>% select(GID,predOf,SELIND) %>% rename(SELIND_full=SELIND),
#           gpreds_medium$gblups[[1]] %>% select(GID,predOf,SELIND) %>% rename(SELIND_medium=SELIND)) %>%
#   group_by(predOf) %>% 
#   summarize(SELIND_corModels=cor(SELIND_full,SELIND_medium))
# predOf  SELIND_corModels
# GEBV	0.9901325			
# GETGV	0.9887326
## EXCELLENT!! :)
```

Choose the best parents for which to predict crosses. Use the GBLUPs from the **full_set** of SNPs.

Take the union of the top 300 clones on the SELIND in terms of GEBV and of GETGV. Probably they will be a very similar list.

```{r select best parents for crossPreds, eval=F}
# SELECT THE BEST PARENTS AS CROSSES-TO-BE-PREDICTED
nParentsToSelect<-300
gpreds_full<-readRDS(file = here::here("output","genomicPredictions_full_set_2021Aug09.rds"))
union_bestGEBVandGETGV<-union(gpreds_full$gblups[[1]] %>% 
                                filter(predOf=="GEBV") %>% 
                                arrange(desc(SELIND)) %>% 
                                slice(1:nParentsToSelect) %$% GID,
                              gpreds_full$gblups[[1]] %>% 
                                filter(predOf=="GETGV") %>% 
                                arrange(desc(SELIND)) %>% 
                                slice(1:nParentsToSelect) %$% GID)
rm(gpreds_full); 
length(union_bestGEBVandGETGV) 
# [1] 365 parents in top nParentsToSelect on SELIND for GEBV/GETGV 

# KEEP ONLY CANDIDATE PARENTS EXPECTED TO BE IN THE FIELD
table(union_bestGEBVandGETGV %in% accessions_infield$FullSampleName)
# FALSE  TRUE 
#   165   200
parentsToPredictCrosses<-union_bestGEBVandGETGV %>% 
  .[. %in% accessions_infield$FullSampleName]
CrossesToPredict<-crosses2predict(parentsToPredictCrosses)
nrow(CrossesToPredict)
# [1] 20100 possible crosses of 200 parents

saveRDS(parentsToPredictCrosses,
        file = here::here("output",
                          "parentsToPredictCrosses_2021Aug10.rds"))
saveRDS(CrossesToPredict,
        file = here::here("output",
                          "CrossesToPredict_2021Aug10.rds"))
```

Predict all pairwise crosses of those 200 parents.

```{r crossPred analysis chunks, eval=F}
# cbsulm17 - 112 cores, 512 GB RAM - 2021 Aug 11 - 8:10am
snpSet<-"full_set"
grms<-list(A=readRDS(here::here("output","kinship_A_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_IITA_2021Aug09.rds")))
predTheMeans<-TRUE; predTheVars<-FALSE
gpreds<-readRDS(file = here::here("output","genomicPredictions_full_set_2021Aug09.rds")); gc()
## takes ~2 minutes, predicting means only

# cbsulm17 - 112 cores, 512 GB RAM - 2021 Aug 10 - 8:25am
snpSet<-"medium_set"
grms<-list(A=readRDS(here::here("output","kinship_A_MediumSNPset_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_MediumSNPset_IITA_2021Aug09.rds")))
predTheMeans<-TRUE; predTheVars<-TRUE
gpreds<-readRDS(file = here::here("output","genomicPredictions_medium_set_2021Aug09.rds")); gc()
# 1263.904

```

```{r run crossPreds for each snpset, eval=F}
snps2keep<-snpsets %>% 
  filter(Set==snpSet) %>% 
  select(snps2keep) %>% 
  unnest(snps2keep)

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()
snpSet; dim(dosages); predTheMeans; predTheVars

start<-proc.time()[3]
crossPreds<-predictCrosses(modelType="DirDom",stdSelInt = 2.0627128, 
                           selInd=TRUE, SIwts=SIwts,
                           CrossesToPredict=CrossesToPredict,
                           snpeffs=gpreds$genomicPredOut[[1]],
                           dosages=dosages,
                           haploMat=haploMat,recombFreqMat=recombFreqMat,
                           ncores=20,nBLASthreads=5,
                           predTheMeans = predTheMeans,
                           predTheVars = predTheVars)
runtime<-proc.time()[3]-start; runtime/60
saveRDS(crossPreds,file = here::here("output",
                                     paste0("genomicMatePredictions_",
                                            snpSet,"_2021Aug10.rds")))

```

## Predict crosses for clones in Ubiaja 

September 10th, 2021: received from I. Kayondo a list of plots in Ubiaja available for crosses (`data/Mate_selection_parental.pool.xlsx`). Match these lines to genomic data and (if not already done) predict matings!

First thing, match the germplasm in the list provided to genotype data...

```{r, eval=F}
library(genomicMateSelectR)
library(tidyverse); library(magrittr)
dbdata<-readRDS(here::here("output","IITA_ExptDesignsDetected_2021Aug08.rds"))
dosages<-readRDS(file=here::here("data","dosages_IITA_2021Aug09.rds"))

parent_candidates<-readxl::read_xlsx(here::here("data","Mate_selection_parental.pool.xlsx")) %>% 
  distinct(accession_name) %>% 
  rename(germplasmName=accession_name) %>% 
  left_join(dbdata %>% 
              distinct(germplasmName,GID,FullSampleName))
```
1354 unique germplasm to select among. Some genotyped multiple times, so 1576 geno matches.

```{r, eval=F}
parent_candidates %>% 
  filter(is.na(GID)) %>% 
  distinct(germplasmName) %>% 
  write.table(.,file=here::here("output","parent_candidates_NOGENOS.txt"),col.names = F, row.names = F, quote = F)

parent_candidates %>% 
  filter(!FullSampleName %in% rownames(dosages),
         !is.na(FullSampleName)) %>% 
  distinct(germplasmName) %>% 
  write.table(.,file=here::here("output","parent_candidates_PEDIGREEREJECT.txt"),col.names = F, row.names = F, quote = F)

parent_candidates %<>% 
  filter(FullSampleName %in% rownames(dosages)) %>% 
  distinct(germplasmName,FullSampleName) %>% 
  group_by(germplasmName) %>% 
  slice(1)
parent_candidates %>% 
    write.table(.,file=here::here("output","parent_candidates_TOPREDICT.txt"),col.names = F, row.names = F, quote = F)

CrossesToPredict<-genomicMateSelectR::crosses2predict(parents)
# nrow(CrossesToPredict)
# [1] 258840 possible crosses of 719 parents
saveRDS(CrossesToPredict,
        file = here::here("output",
                          "CrossesToPredict_2021Sep13.rds"))
```
13 lines don't seem to have matches in SNP data.
552 lines DO have SNP data, but seem to have been excluded from my analyses at the [pedigree-validation step here](03-validatePedigree.html) because they had neither BLUPs (useful training data) nor could their pedigree be validated.

That might not be what you want, since as long as the genomic data matches the clone in the field, even without a verified pedigree it could be a useful parent for crossing. Also, my pedigree verification could have false rejections. So care should be taken. Two lists of clones for you: `output/parent_candidates_NOGENOS.txt`, `output/parent_candidates_PEDIGREEREJECT.txt`.

719 germplasmName have genotype data matches and verified pedigrees. Some have more than one DNA sample, so I will choose one for each. 

Because there are PLENTY of candidate parents, I suggest to proceed for now, but some action or decision should be made about the 552 germplasm not-yet-considered.

File with the list I suggeset to work from: `ouptput/parent_candidates_TOPREDICT.txt`.

You can match the predictions I make with your field plots based on match the "germplasmName" column in this file with the "accession_name" column in the `Mate_selection_parental.pool.xlsx` file you shared with me. 

Regarding the number of crosses I can predict:
It took 21 hrs on a single 112 core server to predict 20100 crosses (pairwise for 200 parents).

For pairwise crosses among 719 in-field parents, 258840 crosses-to-predict. 271 hrs or 11 days. 
I could secure at least 3 servers and do it in about 3-5 actual days. 

Or I could take the top 500 parents (125250 crosses) or 400 parents (80200 crosses) and cut it down to as little as 2 days. 

Delivery end of week vs mid-week?


```{bash, eval=F}
# 1) start a screen shell 
screen; # or screen -r if re-attaching...
# 2) start the singularity Linux shell inside that
singularity shell ~/rocker2.sif; 
# Project directory, so R will use as working dir.
cd /home/mw489/IITA_2021GS/;
# 3) Start R
R
```

```{r, eval=F}
# NEEDED LIBRARIES
require(tidyverse); require(magrittr); library(qs)
library(genomicMateSelectR)

# BLUPs
blups<-readRDS(file=here::here("data","blups_forGP.rds")) %>% 
  dplyr::select(-varcomp)

# DOSAGE MATRIX (UNFILTERED)
dosages<-readRDS(file=here::here("data","dosages_IITA_2021Aug09.rds"))

# RECOMBINATION FREQUENCY MATRIX (UNFILTERED)
recombFreqMat<-qread(file=here::here("data",
                                     "recombFreqMat_1minus2c_2021Aug02.qs"))

# 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_2021Aug09.rds"))


# PARENTS FOR CROSS PREDICTIONS
parents<-read.table(here::here("output","parent_candidates_TOPREDICT.txt"), header = F, stringsAsFactors = F)$V2

# SUBSET HAPS
parenthaps<-sort(c(paste0(parents,"_HapA"),
                   paste0(parents,"_HapB")))
haploMat<-haploMat[parenthaps,]

# SNP SETS TO ANALYZE
snpsets<-readRDS(file = here::here("data","snpsets.rds"))

# 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) 

# CROSSES TO PREDICT
CrossesToPredict<-readRDS(here::here("output",
                                     "CrossesToPredict_2021Sep13.rds"))
```

```{r full_set predMeans, eval=F}
# cbsulm26 - 88 cores, 512 GB RAM - 2021 Sep 13
snpSet<-"full_set"
grms<-list(A=readRDS(here::here("output","kinship_A_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_IITA_2021Aug09.rds")))
predTheMeans<-TRUE; predTheVars<-FALSE
gpreds<-readRDS(file = here::here("output","genomicPredictions_full_set_2021Aug09.rds")); gc()

snps2keep<-snpsets %>% 
  filter(Set==snpSet) %>% 
  select(snps2keep) %>% 
  unnest(snps2keep)

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()
snpSet; dim(dosages); predTheMeans; predTheVars

start<-proc.time()[3]
crossPreds<-predictCrosses(modelType="DirDom",stdSelInt = 2.0627128, 
                           selInd=TRUE, SIwts=SIwts,
                           CrossesToPredict=CrossesToPredict,
                           snpeffs=gpreds$genomicPredOut[[1]],
                           dosages=dosages,
                           haploMat=haploMat,recombFreqMat=recombFreqMat,
                           ncores=17,nBLASthreads=5,
                           predTheMeans = predTheMeans,
                           predTheVars = predTheVars)
runtime<-proc.time()[3]-start; runtime/60
saveRDS(crossPreds,file = here::here("output",
                                     paste0("genomicMatePredictions_",
                                            snpSet,"_2021Sep13.rds")))
# elapsed 
# 14.7936 minutes
```

```{r medium_set predVars in 3 chunks, eval=F}
snpSet<-"medium_set"
grms<-list(A=readRDS(here::here("output","kinship_A_MediumSNPset_IITA_2021Aug09.rds")), 
           D=readRDS(here::here("output","kinship_Dgeno_MediumSNPset_IITA_2021Aug09.rds")))
predTheMeans<-TRUE; predTheVars<-TRUE
gpreds<-readRDS(file = here::here("output","genomicPredictions_medium_set_2021Aug09.rds")); gc()
snps2keep<-snpsets %>% 
  filter(Set==snpSet) %>% 
  select(snps2keep) %>% 
  unnest(snps2keep)

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()
snpSet; dim(dosages); predTheMeans; predTheVars

# DIVIDE CROSSES-TO-PREDICT INTO 3 CHUNKS
CrossesToPredict %<>% 
  mutate(Chunk=rep_along(along = CrossesToPredict$sireID, 1:3) %>% sort(.)) %>% 
  nest(Crosses=c(sireID,damID))

# cbsulm26 - 88 cores, 512 GB RAM - 2021 Sep 13, 10:20am
CrossesToPredict<-CrossesToPredict$Crosses[[1]]
chunk<-1
# [1] "Done predicting fam vars. Took 3867.09 mins for 86280 crosses" = 64.45 hrs = 2.68 days
# [1] "Done predicting fam vars. Took 1658.5 mins for 86280 crosses"
# 5548.574

## two runs of predCrossVars() function internally 
## to predictCrosses() for the DirDom model
## 3867.09/86280 = 0.045 min per cross
## (3867.09*2)/86280 = 0.089 total mins per cross (for DirDom model)

# cbsulm18 - 88 cores, 512 GB RAM - 2021 Sep 13, 9:30am
CrossesToPredict<-CrossesToPredict$Crosses[[2]]
chunk<-2

# [1] "Done predicting fam vars. Took 3700.52 mins for 86280 crosses" = 61.67 hrs = 2.57 days
# [1] "Done predicting fam vars. Took 1600.63 mins for 86280 crosses"
# 5323.687 mins elapsed total

# cbsulm19 - 88 cores, 512 GB RAM - 2021 Sep 13, 9:30am
CrossesToPredict<-CrossesToPredict$Crosses[[3]]
chunk<-3
# [1] "Done predicting fam vars. Took 3659.66 mins for 86280 crosses"
# [1] "Done predicting fam vars. Took 1609.88 mins for 86280 crosses"
# 5292.067 = 88.2 hrs = 3.67 days
CrossesToPredict %>% head

## 128.9 hrs = 5.37 days expected for 3 servers (each 88 core, 512 GB RAM)
## to deliver ~260K cross predictions for an directional dominance model with
## an 8 trait SI and 13K SNPs
```

```{r, eval=F}
start<-proc.time()[3]
crossPreds<-predictCrosses(modelType="DirDom",stdSelInt = 2.0627128, 
                           selInd=TRUE, SIwts=SIwts,
                           CrossesToPredict=CrossesToPredict,
                           snpeffs=gpreds$genomicPredOut[[1]],
                           dosages=dosages,
                           haploMat=haploMat,recombFreqMat=recombFreqMat,
                           ncores=17,nBLASthreads=5,
                           predTheMeans = predTheMeans,
                           predTheVars = predTheVars)
runtime<-proc.time()[3]-start; runtime/60
saveRDS(crossPreds,file = here::here("output",
                                     paste0("genomicMatePredictions_",
                                            snpSet,"_chunk",chunk,"_2021Sep13.rds")))

```



# Write tidy breeder-friendly predictions to disk

## August 2021

Add genetic groups and cohort identifiers and tidy format

```{r output tidy csv of gebv/getgv, eval=F}
library(tidyverse); library(magrittr)
gpreds<-readRDS(file = here::here("output","genomicPredictions_full_set_2021Aug09.rds")) 
gpreds$gblups[[1]] %>% 
  mutate(GeneticGroup=case_when(grepl("2013_|TMS13",GID)~"C1",
                                grepl("TMS14",GID)~"C2",
                                grepl("TMS15",GID)~"C3",
                                grepl("TMS18",GID)~"C4",
                                grepl("TMS20",GID)~"C5",
                                grepl("TMS16",GID)~"TMS16",
                                grepl("TMS17",GID)~"TMS17",
                                grepl("TMS19",GID)~"TMS19",
                                !grepl("2013_|TMS13|TMS14|TMS15|TMS16|TMS17|TMS18|TMS19|TMS20",GID)~"PreGS"),
         #GeneticGroup=factor(GeneticGroup,levels = c("PreGS","C1","C2","C3","C4","C5","TMS16","TMS17","TMS19")),
         Cohort=case_when(grepl("2013_|TMS13",GID)~"TMS13",
                          grepl("TMS14",GID)~"TMS14",
                          grepl("TMS15",GID)~"TMS15",
                          grepl("TMS16",GID)~"TMS16",
                          grepl("TMS17",GID)~"TMS17",
                          grepl("TMS18",GID)~"TMS18",
                          grepl("TMS19",GID)~"TMS19",
                          grepl("TMS20",GID)~"TMS20",
                          !grepl("2013_|TMS13|TMS14|TMS15|TMS16|TMS17|TMS18|TMS19|TMS20",GID)~"PreGS")) %>% 
  relocate(GeneticGroup,.after = "predOf") %>% 
  relocate(Cohort,.after = "GeneticGroup") %>% 
  arrange(predOf,desc(SELIND)) %>% 
  write.csv(.,file = here::here("output","genomicPredictions_full_set_2021Aug09.csv"),
            row.names = F)
```

**NOTE:** For cross predictions, check that the predMean from full and medium set are highly correlated. As long as that is true, combine the predMean from full set with pred var from medium set.

```{r load usefulness predictions, eval=F}
crossPreds_full<-readRDS(file = here::here("output","genomicMatePredictions_full_set_2021Aug10.rds"))
crossPreds_medium<-readRDS(file = here::here("output","genomicMatePredictions_medium_set_2021Aug10.rds"))
crossPreds_medium$rawPreds[[1]]
```


```{r corPredMeans full-medium, eval=F}
crossPreds_full$tidyPreds[[1]] %>% 
  rename(predMean_full=predMean) %>% 
  left_join(crossPreds_medium$tidyPreds[[1]] %>% 
              rename(predMean_medium=predMean)) %>% 
  group_by(predOf,Trait) %>% 
  summarize(corPredMeans=cor(predMean_full,predMean_medium),.groups = 'drop') %>% 
  arrange(desc(corPredMeans)) %$% summary(corPredMeans)
 #   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 # 0.9493  0.9798  0.9928  0.9855  0.9960  0.9981 
```

The lowest corPredMeans was 0.95 for the SELIND, perhaps unsurprisingly. 

Mean corPredMeans=0.9855. I think we are good to go.

Make a plot (below) to examine the scaling of predMean_medium vs. predMean_full to be sure that combining predMean_full with predSD_medium is safe. Seems like it. Everything is on the same scale as expected.
```{r plot predMeans full vs medium, eval=F}
# crossPreds_full$tidyPreds[[1]] %>% 
#   rename(predMean_full=predMean) %>% 
#   left_join(crossPreds_medium$tidyPreds[[1]] %>% 
#               rename(predMean_medium=predMean)) %>% 
#   ggplot(aes(x=predMean_full,y=predMean_medium)) + 
#   geom_point() + 
#   geom_abline(slope=1,color='darkred') +
#   facet_wrap(~Trait, scales='free')
```

Recompute **predUsefulness** using **predMean_full** before saving to disk.

```{r output tidy UC predictions, eval=F}
crossPreds_full$tidyPreds[[1]] %>% 
  left_join(crossPreds_medium$tidyPreds[[1]] %>% 
              rename(predMean_medium=predMean)) %>% 
  mutate(predUsefulness=predMean+(2.0627128*predSD),
         sireGroup=case_when(grepl("2013_|TMS13",sireID)~"TMS13",
                                    grepl("TMS14",sireID)~"TMS14",
                                    grepl("TMS15",sireID)~"TMS15",
                                    grepl("TMS16",sireID)~"TMS16",
                                    grepl("TMS17",sireID)~"TMS17",
                                    grepl("TMS18",sireID)~"TMS18",
                                    grepl("TMS19",sireID)~"TMS19",
                                    grepl("TMS20",sireID)~"TMS20",
                                    !grepl("2013_|TMS13|TMS14|TMS15|TMS16|TMS17|TMS18|TMS19|TMS20",sireID)~"PreGS"),
         damGroup=case_when(grepl("2013_|TMS13",damID)~"TMS13",
                                    grepl("TMS14",damID)~"TMS14",
                                    grepl("TMS15",damID)~"TMS15",
                                    grepl("TMS16",damID)~"TMS16",
                                    grepl("TMS17",damID)~"TMS17",
                                    grepl("TMS18",damID)~"TMS18",
                                    grepl("TMS19",damID)~"TMS19",
                                    grepl("TMS20",damID)~"TMS20",
                            !grepl("2013_|TMS13|TMS14|TMS15|TMS16|TMS17|TMS18|TMS19|TMS20",damID)~"PreGS"),
         CrossGroup=paste0(sireGroup,"x",damGroup)) %>%
  relocate(contains("Group"),.before = "Nsegsnps") %>% 
  relocate(predMean,.before = "predMean_medium") %>% 
  arrange(predOf,Trait,desc(predUsefulness)) %>% 
  write.csv(.,file = here::here("output","genomicMatePredictions_2021Aug10.csv"),
            row.names = F)
```






## September 2021

**NOTE:** For cross predictions, check that the predMean from full and medium set are highly correlated. As long as that is true, combine the predMean from full set with pred var from medium set.

```{r, eval=F}
library(tidyverse); library(magrittr)
crossPreds_full<-readRDS(file = here::here("output","genomicMatePredictions_full_set_2021Sep13.rds"))
crossPreds_medium<-readRDS(file = here::here("output","genomicMatePredictions_medium_set_chunk1_2021Sep13.rds")) %>% 
  bind_rows(readRDS(file = here::here("output","genomicMatePredictions_medium_set_chunk2_2021Sep13.rds"))) %>% 
  bind_rows(readRDS(file = here::here("output","genomicMatePredictions_medium_set_chunk3_2021Sep13.rds"))) %>% 
  select(tidyPreds) %>% 
  unnest(tidyPreds)

```


```{r, eval=F}
crossPreds_full$tidyPreds[[1]] %>% 
  rename(predMean_full=predMean) %>% 
  left_join(crossPreds_medium %>% 
              rename(predMean_medium=predMean)) %>% 
  group_by(predOf,Trait) %>% 
  summarize(corPredMeans=cor(predMean_full,predMean_medium),.groups = 'drop') %>% 
  arrange(desc(corPredMeans)) %$% summary(corPredMeans)
# Joining, by = c("sireID", "damID", "predOf", "Trait")
#    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#  0.9585  0.9732  0.9863  0.9818  0.9889  0.9964 
```

The lowest corPredMeans was 0.95 for the SELIND, perhaps unsurprisingly. 

Mean corPredMeans=0.98. I think we are good to go.

Make a plot (below) to examine the scaling of predMean_medium vs. predMean_full to be sure that combining predMean_full with predSD_medium is safe. Seems like it. Everything is on the same scale as expected.

Recompute **predUsefulness** using **predMean_full** before saving to disk.

```{r, eval=F}
crossPreds_full$tidyPreds[[1]] %>% 
  left_join(crossPreds_medium %>% 
              rename(predMean_medium=predMean)) %>% 
  mutate(predUsefulness=predMean+(2.0627128*predSD),
         sireGroup=case_when(grepl("2013_|TMS13",sireID)~"TMS13",
                                    grepl("TMS14",sireID)~"TMS14",
                                    grepl("TMS15",sireID)~"TMS15",
                                    grepl("TMS16",sireID)~"TMS16",
                                    grepl("TMS17",sireID)~"TMS17",
                                    grepl("TMS18",sireID)~"TMS18",
                                    grepl("TMS19",sireID)~"TMS19",
                                    grepl("TMS20",sireID)~"TMS20",
                                    !grepl("2013_|TMS13|TMS14|TMS15|TMS16|TMS17|TMS18|TMS19|TMS20",sireID)~"PreGS"),
         damGroup=case_when(grepl("2013_|TMS13",damID)~"TMS13",
                                    grepl("TMS14",damID)~"TMS14",
                                    grepl("TMS15",damID)~"TMS15",
                                    grepl("TMS16",damID)~"TMS16",
                                    grepl("TMS17",damID)~"TMS17",
                                    grepl("TMS18",damID)~"TMS18",
                                    grepl("TMS19",damID)~"TMS19",
                                    grepl("TMS20",damID)~"TMS20",
                            !grepl("2013_|TMS13|TMS14|TMS15|TMS16|TMS17|TMS18|TMS19|TMS20",damID)~"PreGS"),
         CrossGroup=paste0(sireGroup,"x",damGroup)) %>%
  relocate(contains("Group"),.before = "Nsegsnps") %>% 
  relocate(predMean,.before = "predMean_medium") %>% 
  arrange(predOf,Trait,desc(predUsefulness)) %>% 
  write.csv(.,file = here::here("output","genomicMatePredictions_2021Sep18.csv"),
            row.names = F)
```





# Results

See [Results](07-Results.html): Home for plots and summary tables.