|
--- |
|
license: mit |
|
tags: |
|
- biology |
|
--- |
|
|
|
Pre-trained single-cell genomics models based on: |
|
- [BarlowTwins](https://proceedings.mlr.press/v139/zbontar21a/zbontar21a.pdf) |
|
- [Bootstrap Your Own Latent](https://papers.nips.cc/paper/2020/file/f3ada80d5c4ee70142b17b8192b2958e-Paper.pdf) |
|
- [Masked Autoencoder](https://openaccess.thecvf.com/content/CVPR2022/papers/He_Masked_Autoencoders_Are_Scalable_Vision_Learners_CVPR_2022_paper.pdf) |
|
- Gene-Program Masked Autoencoder |
|
|
|
Finetuned models for the downstream tasks of: |
|
- Cell Type Prediction |
|
- Gene Expression Reconstruction |
|
- Cross-Modality Prediction (RNA->Proteomics) |
|
- Data Integration |
|
|
|
Training details and adaptations to single-cell data in our project can be found in our paper below. To use the model directly, the same genes must be used in the same order as in the `var.parquet` file. Otherwise, follow the instructions from the repositories below to train a model for custom datasets. |
|
|
|
If you find our work useful, please cite the following paper: |
|
|
|
[**Delineating the Effective Use of Self-Supervised Learning in Single-Cell Genomics**](https://doi.org/10.1101/2024.02.16.580624) |
|
|
|
See also: |
|
|
|
[Repository of the full analysis](https://github.com/theislab/ssl_in_scg) |
|
|
|
[Lean repository for minimal pre-training](https://github.com/theislab/sc_mae) |
