EpGAT

This repository represents an attention-based model named EpGAT for predicting and quantifying post-transcriptional regulatory events using HiC and epigenetic data.

Workflow

Dataset

The data directory before strating the training process, should look like following:

• data
  • <cell_line1_name>
    • as_pred_data (containing quantified AS events for each chromosome)
    • apa_pred_data (containing quantified APA events for each chromosome)
    • eps_data (containing the epigenetics data for each chromosome)
  • <cell_line2_name>
    • as_pred_data
    • apa_pred_data
    • eps_data
  • common_eps.csv (list of the common epigenetic signals' names)
  • <cell_line1_name>.hic ('HiC' file for cell_line1)
  • <cell_line2_name>.hic ('HiC' file for cell_line2)

The python scripts inside preprocess_data folder can be used to prepare the data for application of the model.

Training and testing the model

All the parts of EpGAT model can be trained simultaneously using the python script- 'main.py'. The type of experiment being conducted- cross cell line or cross chromosome should be provided as an argument. Train and test cell lines or chromosomes to be experimented with, should be specified. A number of other options can be also utilized to modify the model parameters. Example command for cross cell line experiment:

python main.py --exp_type cross_cell_line --train_cell_line MCF7 --test_cell_line K562 --chromosome 1 --event as --root_directory data

Example command for cross chromosome experiment:

python main.py --exp_type cross_chromosome --cell_line MCF7 --train_chrs 1 2 3 4 5 --val_chrs 6 13 --test_chrs 7 14 --event apa --root_directory data