https://github.com/dohlee/bpnet-pytorch

Implementation of BPNet, a base-resolution convolutional neural network for transcription-factor binding prediction, in PyTorch.
https://github.com/dohlee/bpnet-pytorch

bioinformatics biology computational-biology deep-learning reproduction reproduction-code transcription-factor-binding

Last synced: about 2 months ago
JSON representation

Implementation of BPNet, a base-resolution convolutional neural network for transcription-factor binding prediction, in PyTorch.

• Host: GitHub
• URL: https://github.com/dohlee/bpnet-pytorch
• Owner: dohlee
• Created: 2023-02-20T14:52:13.000Z (almost 2 years ago)
• Default Branch: master
• Last Pushed: 2023-05-07T11:32:38.000Z (over 1 year ago)
• Last Synced: 2024-11-07T02:42:56.678Z (about 2 months ago)
• Topics: bioinformatics, biology, computational-biology, deep-learning, reproduction, reproduction-code, transcription-factor-binding
• Language: Python
• Homepage:
• Size: 83 KB
• Stars: 1
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# bpnet-pytorch (wip)

![model](img/banner.png)

Implementation of BPNet, a base-resolution deep neural network for functional genomics tasks. The offical implementation of BPNet can be found [here](https://github.com/kundajelab/bpnet).

## Installation

```shell

$ pip install bpnet-pytorch

```

## Usage

```Python

from bpnet_pytorch import BPNet

model = BPNet()

x = torch.randn(1, 4, 1000)

out = model(x)

# out['x'] contains the output of the convolution layers.

# May not be useful, but left for debugging purpose for now.

### shape: (1, 64, 1000)

# out['profile'] contains the output of profile head.

### shape: (1, 1000, 2), 2 for +/- strands.

# out['total_count'] contains the output of total count head.

### shape: (1, 2), 2 for +/- strands.

```

## Training

### Data

Training data used to train the original BPNet model can be found [here](https://zenodo.org/record/3371216) (30.8GB).

## TODO

- [x] Confirm that the receptive field is +-1034bp.

![receptive-field-check](img/receptive_field_check.png)

- [x] Implement early stopping.

- [x] Implement multinomial negative log-likelihood loss.

- [ ] Modify heads for multi-task (multi-TF) prediction.

- [ ] Prepare training data.

- [ ] Train the model and reproduce the performance.

## Citation

```bibtex

@article{avsec2021base,

  title={Base-resolution models of transcription-factor binding reveal soft motif syntax},

  author={Avsec, {\v{Z}}iga and Weilert, Melanie and Shrikumar, 

    Avanti and Krueger, Sabrina and Alexandari, Amr and Dalal, Khyati and Fropf,

    Robin and McAnany, Charles and Gagneur, Julien and Kundaje, Anshul and others},

  journal={Nature Genetics},

  volume={53},

  number={3},

  pages={354--366},

  year={2021},

  publisher={Nature Publishing Group US New York}

}

```