Home

Welcome to the MethylCapsNet wiki!

Currently, the wiki is being refactored. Stay tuned for updates! You can find our arxiv here: https://www.biorxiv.org/content/10.1101/2020.08.14.251306v1

Experimental Docker build command:

docker build docker/ -t methylcapsnet:0.1
docker run methylcapsnet:0.1
methylcaps-model test_pipeline

Basic Run Commands:

methylcaps-hypscan hyperparameter_scan -ic disease_only -gpu -t -j 2 -a "conda activate methylnet" -u -tt 120 -rs 42 -ne 1 -s random
methylcaps-hypscan hyperparameter_scan -ic disease_only -gpu -t -j 2 -a "conda activate methylnet" -u -tt 120 -rs 42 -ne 1 -s random --retrain_top_job
methylcaps-hypscan hyperparameter_scan -ic disease_only -gpu -t -j 2 -a "conda activate methylnet" -u -tt 120 -rs 42 -ne 1 -s random --retrain_top_job --output_top_job_params
methylcaps-model execute_top_job

Example run of custom parameters:

CUDA_VISIBLE_DEVICES=0 methylcaps-model model_capsnet --bin_len 900000 --caps_out_len 15 --gamma 0.0001 --learning_rate 0.001 --min_capsule_len 300 --primary_caps_out_len 35 --routing_iterations 2 --hidden_topology 160,40 --decoder_topology 220,270 --train_methyl_array ./train_val_test_sets/train_methyl_array.pkl --val_methyl_array ./train_val_test_sets/val_methyl_array.pkl --interest_col disease_only --custom_loss none --job 6423388 --batch_size 16 --n_epochs 1 --capsule_choice genomic_binned --gamma2 0.01  --predict
methylcaps-model model_capsnet --bin_len 900000 --caps_out_len 15 --gamma 0.0001 --learning_rate 0.001 --min_capsule_len 300 --primary_caps_out_len 35 --routing_iterations 2 --hidden_topology 160,40 --decoder_topology 220,270 --train_methyl_array ./train_val_test_sets/train_methyl_array.pkl --val_methyl_array ./train_val_test_sets/val_methyl_array.pkl --interest_col disease_only --custom_loss none --job 6423388 --batch_size 16 --n_epochs 1 --capsule_choice genomic_binned --gamma2 0.01  --predict

Then, after receiving prediction file, launch web app at web_app/WebApp.ipynb via jupyter notebook. Alternatively, use jupyter to analyze prediction file and process outputs using downstream R programs such as Circos, WGCNA, limma, etc... See example_R_scripts .