CLARINET

This code is implementation of our paper [CLARINET].
CLARINET is an attention based deep learning model that predicts whether a TF binds to a given DNA sequence.
CLARINET takes DNA sequence as one-hot vector and predicts 690 TF binding tasks.
The above picture is an overview of our model.

Requirements

• pytorch
• keras

Usage

Training model

0. Download data (http://deepsea.princeton.edu/media/code/deepsea_train_bundle.v0.9.tar.gz)
   • Data includes train.mat, valid.mat, and test.mat
   • Place that data at the ./data folder
1. Train model and save trained model to ./model folder
   • python clarinet.py train ./data ./model/[model_name]
2. Evaluate trained model
   • python clarinet.py eval ./data ./model/[model_name]

Attention analysis

python attn/analysis.py [] []

SNP prioritization

python snp/analysis.py [] []

Result

Model prediction result

ROC AUC scatter plot

Attention analysis result

Heatmap plot

SNP prioritization result

ROC AUC plot