6.8611 Final Project: Cell Type Specific DNA Language Models

Section 1: Vanilla Input Cell Type Specific Model Training

Description: Training code for our cell type specific (non architectural changes) version of the Enformer model from Žiga Avsec et al., 2021

Found In:/Cell_Type_Specific_Training

Contents:

1. Train_Model_OG.sh: bash script submitting batch job to slurm scheduler on compute cluster to run Train_Model_OG.py
2. Train_Model_OG.py: python file containing training code

Section 2: Autoencoder Head Cell Type Specific Model Training

Description: Training code for our autoencoder headed cell type specific version of the Enformer model from Žiga Avsec et al., 2021

Found In:/AutoEnc-CellTypeSpec_Traiing

Contents:

1. /Autoencoder_Training: Training scripts for the autoencoder head, trained on reconstruction loss a. Train_Autoencoder.sh: Batch job submission for python file b. Train_Autoencoder_v2.py: Python file training autoencoder head on reconstruction loss
2. /Full_Model_Training: Training scripts for the full model including reconstruction loss, trained on expression prediction a. Train_Full_Model.sh: Batch job submission for python file b. Train_Full_Model.py: Python script training full model, including pretrained autoencoder on gene expression prediction

Section 3: Evaluation

Found In: /pearson_correlation_code.ipynb

Contents:

Code to produce evaluation plots for trained models