Stress Predictor based on Machine Learning

Learning

With prepared dataset,

python learn.py dataset/dataset_learn_v2_005.limma.csv dataset/dataset_learn_v2_005.label.csv --reduce_updown --save output/model_v2 --save_csv output/model_v2 --epoch_count 1000 --batch_size 0

• --reduce_updown Use this parameter if dataset is separated into up and down column.
• --save Directory to save model data
• --save_csv Directory to save model data (in csv form)
• --epoch_count epoch to learn
• --batch_size learning batch size

Testing prediction /w learnt parameter

python test.py dataset/dataset_learn_v2_005.limma.csv dataset/dataset_learn_v2_005.label.csv --load output/model_v2 --save output/model_v2_test
python test.py dataset/dataset_test_005.limma.csv dataset/dataset_test_005.label.csv --load output/model_v2 --save output/model_v2_test

• --load Directory of learn.py output
• --save Directory to save test result

GSEA (Gene set enrichment test) Testing

python GOenrichment.py $1/output/params_generator.csv --trait2genes GOBPname2gene.arabidopsis.txt --column_name "heat,salt,drought,cold" --count_cut 500 -o $1/output/GSEA_learn_top500.csv --descending --max_trait_cut 1

• --trait2genes Path to trait-to-gene file.
• --column_name Column names including p-value.
• --count_cut How many gene set to be shown?
• -o Path to output
• --descending Result data in descending p-value order
• --max-trait-cut Remove genes with too little traits.

How to generate dataset?

Just prepare your expression matrix (CEL file processed) and metadata (may write TSD header or generate from CSV). then use these commands:

# split total dataset file into time-series-data formatted file
python ../dataset.py tsd -f cold.csv
python ../dataset.py tsd -f heat.csv
# read directory just for test
python ../dataset.py read -d ./

# without any processing, just output raw p-value
python ../dataset.py compile -d ./ --tool limma -t dataset_test_005
# processing p-value /w threshold .05, (signed)
python ../dataset.py compile -d ./ --tool limma -t dataset_test_005 --pvalue 0.05
# processing p-value /w threshold .05, with reindexing & separating up/down signal (unsigned)
python ../dataset.py compile -d ./ --tool limma -t dataset_test_005 --pvalue 0.05 --updown --reindex_df ../dataprocess/ttest_pval.csv

# generate gene list order for comparison with other result (optional)
python dataset.py gen_genelist -f data/GSE3326_1.tsd -t dataset/index
# create dataset with pvalue 0.05
python dataset.py compile -d data_test/ --tool limma -t dataset/dataset_test_005 --pvalue 0.05 --reindex_file dataset/index.txt --updown --label_index "heat,salt,drought,cold"
# create dataset with additional filter
python dataset.py compile -d data/ --tool limma -t dataset/dataset_learn_v2_005 --pvalue 0.05 --reindex_file dataset/index.txt  --updown --label_index "heat,salt,drought,cold" --filters "Species:Arabidopsis Thaliana,MinRepCnt:2"

Then DEG infomation will be stored at dataset_test.csv matrix. You can use --filter to select TSData with specific condition (with column name and value).