Basset

Deep convolutional neural network for predicting chromatin accessibility (DNase I hypersensitivity) from DNA sequence.

Disclaimer

This is an UNOFFICIAL implementation of Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks by David R. Kelley et al.

The OFFICIAL repository of Basset is at davek44/Basset.

The MultiMolecule team has confirmed that the provided model and checkpoints are producing the same intermediate representations as the original implementation.

The team releasing Basset did not write this model card for this model so this model card has been written by the MultiMolecule team.

Model Details

Basset is a convolutional neural network (CNN) trained to predict the chromatin accessibility (DNase I hypersensitivity) of a DNA sequence across 164 cell types. The model consumes a fixed-length 600 bp one-hot encoded DNA sequence and applies three convolutional blocks (convolution, batch normalization, ReLU, and max pooling) followed by two fully-connected blocks before a multi-label binary classification head. Please refer to the Training Details section for more information on the training process.

Model Specification

• Architecture: 3 convolutional layers + 2 fully-connected layers
• Convolution filters: 300, 200, 200
• Convolution kernel sizes: 19, 11, 7
• Max-pool sizes: 3, 4, 4
• Fully-connected sizes: 1000, 1000
• Input length: 600 bp
• Number of labels: 164 (DNase I hypersensitivity, multi-label binary)

Num Conv Layers	Num FC Layers	Hidden Size	Num Parameters (M)	FLOPs (G)	MACs (G)	Max Num Tokens
3	2	1000	4.14	0.30	0.15	600

Links

• Code: multimolecule.basset
• Weights: multimolecule/basset
• Paper: Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks
• Developed by: David R. Kelley, Jasper Snoek, John L. Rinn
• Original Repository: davek44/Basset

Usage

The model file depends on the multimolecule library. You can install it using pip:

pip install multimolecule

Direct Use

Chromatin Accessibility Prediction

You can use this model directly to predict the DNase I hypersensitivity of a DNA sequence:

>>> import torch
>>> from multimolecule import DnaTokenizer, BassetForSequencePrediction

>>> tokenizer = DnaTokenizer.from_pretrained("multimolecule/basset")
>>> model = BassetForSequencePrediction.from_pretrained("multimolecule/basset")
>>> input = tokenizer("ACGT" * 150, return_tensors="pt")
>>> output = model(**input)

>>> output.logits.shape
torch.Size([1, 164])

Training Details

Basset was trained to predict the chromatin accessibility of DNA sequences across a panel of cell types.

Training Data

Basset was trained on DNase I hypersensitivity peaks from ENCODE and the Roadmap Epigenomics project, covering 164 cell types. Each 600 bp genomic interval is labeled with a binary vector indicating which of the 164 cell types show an accessibility peak overlapping that interval.

Training Procedure

Pre-training

The model was trained to minimize a multi-label binary cross-entropy loss, comparing its predicted per-cell-type accessibility probabilities against the observed DNase I hypersensitivity labels.

• Optimizer: RMSprop
• Loss: Multi-label binary cross-entropy
• Regularization: Batch normalization and dropout

Citation

@article{kelley2016basset,
  author    = {Kelley, David R. and Snoek, Jasper and Rinn, John L.},
  title     = {Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks},
  journal   = {Genome Research},
  volume    = 26,
  number    = 7,
  pages     = {990--999},
  year      = 2016,
  publisher = {Cold Spring Harbor Laboratory Press},
  doi       = {10.1101/gr.200535.115}
}

The artifacts distributed in this repository are part of the MultiMolecule project. If you use MultiMolecule in your research, you must cite the MultiMolecule project as follows:

@software{chen_2024_12638419,
  author    = {Chen, Zhiyuan and Zhu, Sophia Y.},
  title     = {MultiMolecule},
  doi       = {10.5281/zenodo.12638419},
  publisher = {Zenodo},
  url       = {https://doi.org/10.5281/zenodo.12638419},
  year      = 2024,
  month     = may,
  day       = 4
}

Contact

Please use GitHub issues of MultiMolecule for any questions or comments on the model card.

Please contact the authors of the Basset paper for questions or comments on the paper/model.

License

This model implementation is licensed under the GNU Affero General Public License.

For additional terms and clarifications, please refer to our License FAQ.

SPDX-License-Identifier: AGPL-3.0-or-later