TECH-3D & eTECH-3D

This repository contains our implementation for chromatin 3D inference from Hi-C data using deep learning. A detailed description of the project can be found in root -> report -> written_report -> master_thesis_report.pdf.

Repository Organisation

• tech3d: contains the implementation of TECH-3D (Transfer learning Encoder for Chromatin 3D structure prediction) for different encoders (Linear or Bi-LSTM) and train datasets (Synthetic Random Walk Structures or Synthetic Biological Structures). To run the code, one need to:

  1. Generate the synthetic train data by running the appropriate data generator (e.g. data_generation -> synthetic_random_trussart_data_generation.ipynb).
  2. Train the network on the previously generated data by running the appropriate notebook (e.g. networks -> synthetic_random_trussart -> linear -> synthetic_random_trussart_linear.ipynb).

  In the example above, TECH-3D is designed the for test Trussart Hi-C matrix at 5Kb, i.e. the generated synthetic structures and contact matrices have 202 bins. A similar implementation for other test datasets (GEM12878 and Fission Yeast) is also available.

• etech3d: contains the implementation of eTECH-3D (ensemble-based Transfer learning Encoder for Chromatin 3D population of structures analysis). As for TECH-3D, eTECH-3D uses transfer learning and the working pipeline is similar as above.

• experiments: contains the experiments evaluating both TECH-3D and eTECH-3D as well as benchmark algorithms (REACH-3D, GEM and miniMDS). All the experimental figures of the report can be regenerated from the jupyter notebooks in this folder.

• previous_works: contains the implementations of REACH-3D, GEM and miniMDS as well as data formatters designed to transform the Hi-C contact map (e.g. the Trussart contact map at 5Kb) into the required format. For GEM and miniMDS, the implementation is obtained from the official repository. On the other hand, REACH-3D was developed from scratch in Pytorch 1.9.0 (originally in Tensorflow 1.9.0).

• report.zip: contains all the literature covering TECH-3D and eTECH-3D.

Installation

First, clone the repository. Then, we recommend to work in a separate python environment. For this, run the following command at the root (use the package manager pip to install python venv, if required):

python3 -m virtualenv tech3d_venv
source tech3d_venv/bin/activate

Finally, install the required libraries:

pip3 install -r requirements.txt

Notes

• In the code, it is possible that the name tREACH-3D is used instead of TECH-3D. This is because the name of our framework changed lately. Similarly, eTECH-3D can be referred as tCev-3D.
• In the repository, data files and external repositories are compressed for memory space efficiency. We recommend to unzip the files before working on the code.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT