All scripts for this project on HPCC: /mnt/home/azodichr/GitHub/GenomicPrediction_2018/scripts/

Run scripts from a directory that has directories (named with the project ID) for each dataset. Datasets for this project are listed below. Each directory should contain a subdirectory 01_Data that contains the geno.csv and pheno.csv file formatted as described in Step #1.

Project IDs: -rice_DP_Spindel

-sorgh_DP_Fernan

-soy_NAM_xavier

-spruce_Dial_beaulieu

-swgrs_DP_Lipka # Still working on genotype data

Project_ID/: 00_RawData/ 01_Data/ pheno.csv geno.csv

Since all input data so far has been slightly different, this needs to be done manually script: scripts/data_preprocessing.R

Create two files with matching indexes: geno.csv, pheno.csv

-Remove lines missing any phenotypes of interest

-Remove lines with missing genotype information

-Take the average of phenotypes across years and replicates

-Convert genotype data to [-1,0,1] format corresponding to [aa, Aa, AA]

 python make_CVs.py -id [ID] 

How well can you predict a trait value using just the population structure? These scripts run principle component analysis on the SNP data, then use the top N PCs to predict the trait values.

Rscript getPCs.R [ID]
Rscript predict_PC.R [ID] [N]  

Example qsub file for submitting to HPCC as an array job: /mnt/home/azodichr/03_GenomicSelection/qsub_files/qsub_rrBLUP.txt

*Be sure to change the ID to your project ID!

 Rscript predict_rrBLUP.R [ID] [cv_num] [trait/all] [path/to/output/dir] 

Using the BGLR R package. Bayesian Methods available: BayesA, BayesB, B-LASSO, and B-Ridge Regression Example qsub file for submitting to HPCC as an array job: /mnt/home/azodichr/03_GenomicSelection/qsub_files/qsub_BayA.txt

 Rscript predict_BayesA.R [ID] [cv_num] [optional_parameter] [trait/all] [path/to/output/dir]

-BayesA: ID - CV_num - deg.freedom - trait - path_to_output

-BayesB: ID - CV_num - deg.freedom - trait - path_to_output

-BRR: ID - CV_num - trait - path_to_output

-BL: ID - CV_num - trait - path_to_output

Utilizes the Shiu Lab machine learning pipeline (https://github.com/ShiuLab/ML-Pipeline) implementing ML with SciKit-Learn (http://scikit-learn.org/stable/)

Algorithms available for regression: RF (random forest), SVM (support vector machine), GB (gradient boosting), and LogReg (logistic regression)

See Shiu Lab ML-Pipeline repository for environment requirements. If working on MSU's HPCC, the environment is available at:

export PATH=/mnt/home/azodichr/miniconda3/bin:$PATH

Example run:

python python /mnt/home/azodichr/GitHub/ML-Pipeline/ML_regression.py -df geno.csv -df_Y pheno.csv,TRAIT -cv_set CVFs.csv  -sep , -alg [ALG] -gs T -plots F -p 5 -save SAVE_NAME -out PATH/TO/OUTPUT

python ~/GitHub/TF-GenomicSelection/make_jobs_tf.py job_header.txt test_params.txt for i in job*.sh; do qsub $i; done

Get the average accuracies: ** modify line 14 to include the results you are interested in $ python pull_results.py Output: RESULTS.csv

Plots the results pulled above! $ Rscript plot_results.R Output: RESULTS.pdf

-rrBLUP: no parameters to define

-Bayes A: degrees of freedom (2, 5, 7, 15)

-Bayes B: degrees of freedom (2, 5, 7, 15)

-B-LASSO: no parameters to define

-B-RR: no parameters to define

-SVM: Kernel (linear, polynomial, rbf), C (0.01, 0.1, 0.5, 1, 10, 50, 100), gamma (10, 1, 0.1, 0.01, 0.001, 0.0001, 0.00001)

-RF: Max Depth (3, 5, 10, 50), Max Features (10%, 50%, 100%, sqrt(#feat), log2(#feat))

-GB: Learning Rate (0.0001, 0.001, 0.01, 0.1, 1), Max Depth (3, 5, 10, 50), Max Features (10%, 50%, 100%, sqrt(#feat), log2(#feat))

-ANN: Activation function (ReLU, sigmoid), Number of hidden layers (1, 2, 3), Largest Node Size (10, 100, 1000), Regularization (l1, l2, both)

Before we were breaking a ML rule by doing feature selection on the whole data set. Try a side experiment with FS using BayesA, LASSO, RF, and Relief (top 10, 100, 250, 500, 750, 1000, 1500, 2000). Using cv_1-cv_10, do FS and build the models on groups 1-4, then test the fs model on group 5.

These FS methods are built into the Shiu Lab's Feature_Selection.py script. I made small changes so that the FS would be done on folds 1-4 of the CV set and changes to the ML_regression.py script allow for it to build the model on that same set and apply it to the cv5.

Example run: (Will use RF to do feature selection on height (HT), selecting the top 10 features for the first 10 cv_reps.

python ~/GitHub/ML-Pipeline/Feature_Selection.py -df geno.csv -df_class pheno.csv,HT -f RF -sep ',' -n 10 -CVs CVFs.csv -reps 10 -type r -list T -save geno_HT_RF_10
python ~shius/codes/qsub_hpc.py -f submit -u azodichr -c runcc_FS_RF_Relief.txt -w 239 -m 10 -wd /mnt/home/azodichr/03_GenomicSelection/sorgh_DP_Fernan/01_Data

Modified the predict_BayesA.R script to do FS on just cv1-4. Rscript FS_BayesA.R maize_DP_Crossa/ Rscript FS_BayesA.R swgrs_DP_Lipka/ Rscript FS_BayesA.R soy_NAM_xavier/

python ~shius/codes/qsub_hpc.py -f submit -u zhan1165 -w 239 -m 40 -wd /mnt/home/zhan1165/azodi/maize/01_Data/ -c runcc_bayesAFS.txt

Run models with selected features

#maize_DP_Crossa for i in /fl_f; do Rscript ../predict_BayesA.R maize_DP_Crossa $i fl_f; done for i in 02_LASSO/maize_flf; do Rscript ../predict_BayesA.R maize_DP_Crossa $i fl_f; done for i in /fl_m; do Rscript ../predict_BayesA.R maize_DP_Crossa $i fl_m; done for i in 02_LASSO/maize_flm; do Rscript ../predict_BayesA.R maize_DP_Crossa $i fl_m; done for i in */GY; do Rscript ../predict_BayesA.R maize_DP_Crossa $i GY; done #swgrs_DP_Lipka for i in */Plan; do Rscript ../predict_BayesA.R swgrs_DP_Lipka $i Plant_Height; done for i in */Anthes; do Rscript ../predict_BayesA.R swgrs_DP_Lipka $i Anthesis_Date_8632; done for i in */Standability; do Rscript ../predict_BayesA.R swgrs_DP_Lipka $i Standability; done #soy_NAM_xavier for i in */height; do Rscript ../predict_BayesA.R soy_NAM_xavier $i height; done for i in */R8; do Rscript ../predict_BayesA.R soy_NAM_xavier $i R8; done for i in */yield; do Rscript ../predict_BayesA.R soy_NAM_xavier $i yield; done

export R_LIBS_USER=~/R/library #maize_DP_Crossa for i in /fl_f; do Rscript ../predict_rrBLUP.R maize_DP_Crossa $i fl_f; done for i in 02_LASSO/maize_flf; do Rscript ../predict_rrBLUP.R maize_DP_Crossa $i fl_f; done for i in /fl_m; do Rscript ../predict_rrBLUP.R maize_DP_Crossa $i fl_m; done for i in 02_LASSO/maize_flm; do Rscript ../predict_rrBLUP.R maize_DP_Crossa $i fl_f; done for i in */GY; do Rscript ../predict_rrBLUP.R maize_DP_Crossa $i GY; done #swgrs_DP_Lipka for i in /Plant_Height; do Rscript ../predict_rrBLUP.R swgrs_DP_Lipka $i Plant_Height; done for i in 02_LASSO/swgrs_Pla; do Rscript ../predict_rrBLUP.R swgrs_DP_Lipka $i Plant_Height; done for i in /Anthesis_Date_8632; do Rscript ../predict_rrBLUP.R swgrs_DP_Lipka $i Anthesis_Date_8632; done for i in 02_LASSO/swgrs_AnthesisDate_; do Rscript ../predict_rrBLUP.R swgrs_DP_Lipka $i Anthesis_Date_8632; done for i in */Standability; do Rscript ../predict_rrBLUP.R swgrs_DP_Lipka $i Standability; done #soy_NAM_xavier python ~shius/codes/qsub_hpc.py -f submit -u azodichr -c run_rrBLUP.txt -w 230 -m 10 -n 200 -wd /mnt/home/azodichr/03_GenomicSelection/02_FeatureSelection/soy_NAM_xavier/

#maize_DP_Crossa for i in */fl_f; do python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.txt,fl_f -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag maize_DP_Crossa -feat $i; done for i in */fl_m; do python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.txt,fl_m -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag maize_DP_Crossa -feat $i; done for i in */GY; do python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.txt,GY -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag maize_DP_Crossa -feat $i; done #swgrs_DP_Lipka for i in */Plant_Height; do python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.csv,Plant_Height -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag swgrs_DP_Lipka -feat $i; done for i in */Anthesis_Date_8632; do python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.csv,Anthesis_Date_8632 -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag swgrs_DP_Lipka -feat $i; done for i in */Standability; do python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.csv,Standability -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag swgrs_DP_Lipka -feat $i; done #soy_NAM_xavier for i in */height; do echo python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.txt,height -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag soy_NAM_xavier -feat $i; done for i in */R8; do echo python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.txt,R8 -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag soy_NAM_xavier -feat $i; done for i in */yield; do echo python ../ML_regression.py -df geno.csv -df_Y pheno_cv1_NAs.txt,yield -gs T -n 5 -sep ',' -alg RF -UNKNOWN ? -answers pheno.csv -tag soy_NAM_xavier -feat $i; done

#maize_DP_Crossa for i in */fl_f; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label fl_f -tag maize_DP_Crossa; done for i in */fl_m; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label fl_m -tag maize_DP_Crossa; done for i in /GY; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label GY -tag maize_DP_Crossa; done python ~/GitHub/TF-GenomicSelection/make_jobs_tf.py header_jobs.txt run_MLP.txt for i in job.sh; do qsub $i; done

#swgrs_DP_Lipka for i in */Plant_Height; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label Plant_Height -tag swgrs_DP_Lipka; done for i in */Anthesis_Date_8632; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label Anthesis_Date_8632 -tag swgrs_DP_Lipka; done for i in /Standability; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label Standability -tag swgrs_DP_Lipka; done python ~/GitHub/TF-GenomicSelection/make_jobs_tf.py header_jobs.txt run_MLP.txt for i in job.sh; do qsub $i; done

#soy_NAM_xavier for i in */height; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label height -tag soy_NAM_xavier; done for i in */R8; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label R8 -tag soy_NAM_xavier; done for i in /yield; do echo python ../TF_MLP_GridSearch.py -x geno.csv -y pheno.csv -cv CVFs.csv -feat $i -label yield -tag soy_NAM_xavier; done python ~/GitHub/TF-GenomicSelection/make_jobs_tf.py header_jobs.txt run_MLP.txt for i in job.sh; do qsub $i; done

python compare_predictions.py -d maize_DP_Crossa -p 0.2 python compare_predictions.py -d maize_DP_Crossa -p 0.1 python compare_predictions.py -d rice_DP_Spindel -p 0.2 python compare_predictions.py -d rice_DP_Spindel -p 0.1 python compare_predictions.py -d soy_NAM_xavier -p 0.2 python compare_predictions.py -d soy_NAM_xavier -p 0.1 python compare_predictions.py -d spruce_Dial_beaulieu -p 0.2 python compare_predictions.py -d spruce_Dial_beaulieu -p 0.1 python compare_predictions.py -d swgrs_DP_Lipka -p 0.2 python compare_predictions.py -d swgrs_DP_Lipka -p 0.1

cat maize_DP_Crossa/topOvlp_0.1.txt rice_DP_Spindel/topOvlp_0.1.txt soy_NAM_xavier/topOvlp_0.1.txt spruce_Dial_beaulieu/topOvlp_0.1.txt swgrs_DP_Lipka/topOvlp_0.1.txt > topOvlp_all_0.1.txt

cat maize_DP_Crossa/topOvlp_0.2.txt rice_DP_Spindel/topOvlp_0.2.txt soy_NAM_xavier/topOvlp_0.2.txt spruce_Dial_beaulieu/topOvlp_0.2.txt swgrs_DP_Lipka/topOvlp_0.2.txt > topOvlp_all_0.2.txt

Rscript -e "install.packages('rrBLUP', lib='~/R/library', contriburl=contrib.url('http://cran.r-project.org/'))"

Modified the predict_BayesA.R script to do FS on just cv1-4. Rscript FS_BayesA.R maize_DP_Crossa/ Rscript FS_BayesA.R swgrs_DP_Lipka/ Rscript FS_BayesA.R soy_NAM_xavier/

RandomForest: See run files: 01_RF/run_FS_RF.txt Example: $ python ~/GitHub/ML-Pipeline/Feature_Selection.py -f RF -df ../geno.csv -df_class ../pheno_cv1_NAs.txt,height -n 1500 -type r -ignore '?' -sep ',' -list T -save soy_height_RF_1500

LASSO: python ~/GitHub/ML-Pipeline/Feature_Selection.py -f LASSO -df ~/03_GenomicSelection/02_FeatureSelection/maize_DP_Crossa/geno.csv -df_class ~/03_GenomicSelection/02_FeatureSelection/maize_DP_Crossa/pheno_cv1_NAs.txt,GY -sep ',' -list T -type r -ignore '?' -p 0.01 python ~/GitHub/ML-Pipeline/Feature_Selection.py -f LASSO -df ~/03_GenomicSelection/02_FeatureSelection/swgrs_DP_Lipka/geno.csv -df_class ~/03_GenomicSelection/02_FeatureSelection/swgrs_DP_Lipka/pheno_cv1_NAs.csv,Plant_Height -sep ',' -list T -type r -ignore '?' -p 0.45 python ~/GitHub/ML-Pipeline/Feature_Selection.py -f LASSO -df ~/03_GenomicSelection/02_FeatureSelection/soy_NAM_xavier/geno.csv -df_class ~/03_GenomicSelection/02_FeatureSelection/soy_NAM_xavier/pheno_cv1_NAs.txt,height -sep ',' -list T -type r -ignore '?' -p 1.03

Relief: python ~shius/codes/qsub_hpc.py -f submit -u azodichr -c run_FS_relief.txt -w 20 -m 50 -wd /mnt/home/azodichr/03_GenomicSelection/02_FeatureSelection/maize_DP_Crossa/03_relief/ python ~shius/codes/qsub_hpc.py -f submit -u azodichr -c run_FS_relief.txt -w 20 -m 50 -wd /mnt/home/azodichr/03_GenomicSelection/02_FeatureSelection/swgrs_DP_Lipka/03_relief/ python ~shius/codes/qsub_hpc.py -f submit -u azodichr -c run_FS_relief.txt -w 230 -m 80 -wd /mnt/home/azodichr/03_GenomicSelection/02_FeatureSelection/soy_NAM_xavier/03_relief/