Implementation of experiments incorporating projection layer to simplify representation of motifs and improve performance in neural networks for regulatory genomics described in Projection layers improve deep learning models of regulatory DNA function (bioRxiv)

python scripts/run_experiments.py <experiment_name>

Experiment names with associated settings are defined in expset_settings:

• expset_settings/500bp_slim.py - experiments with projection layer, dropout in 3 layer cnn with varying number of filters in first layer
• expset_settings/final_crnn_slimpy - experiments with convolutional-recurrent neural networks using projection layer to improve on DanQ performance on DeepSEA dataset.

e.g. to run the convolutional recurrent model with 320 first layer filters

python scripts/run_experiments.py final_crnn_320_300rd20

This project is licensed under the terms of the MIT license