scgpt
Towards Building a Foundation Model for Single-Cell Multi-omics Using Generative AI
Generative pre-trained models have achieved remarkable success in various domains such as natural language processing and computer vision. Specifically, the combination of large-scale diverse datasets and pre-trained transformers has emerged as a promising approach for developing foundation models. Drawing parallels between linguistic constructs and cellular biology - where texts comprise words, similarly, cells are defined by genes - our study probes the applicability of foundation models to advance cellular biology and genetics research. Utilizing the burgeoning single-cell sequencing data, we have pioneered the construction of a foundation model for single-cell biology, scGPT, which is based on generative pre-trained transformer across a repository of over 33 million cells. Our findings illustrate that scGPT, a generative pre-trained transformer, effectively distills critical biological insights concerning genes and cells. Through the further adaptation of transfer learning, scGPT can be optimized to achieve superior performance across diverse downstream applications. This includes tasks such as cell-type annotation, multi-batch integration, multi-omic integration, genetic perturbation prediction, and gene network inference.
Haotian Cui , Chloe Wang , Hassaan Maan , Kuan Pang , Fengning Luo , Bo Wang ,
Monthly downloads
1000
scGPT ¶
This is the official codebase for scGPT: Towards Building a Foundation Model for Single-Cell Multi-omics Using Generative AI.
!UPDATE: We have released several new pretrained scGPT checkpoints. Please see the Pretrained scGPT checkpoints section for more details.
Installation ¶
scGPT is available on PyPI. To install scGPT, run the following command:
$ pip install scgpt
[Optional] We recommend using wandb for logging and visualization.
$ pip install wandb
For developing, we are using the Poetry package manager. To install Poetry, follow the instructions here.
$ git clone this-repo-url
$ cd scGPT
$ poetry install
Note: The flash-attn dependency usually requires specific GPU and CUDA version. If you encounter any issues, please refer to the flash-attn repository for installation instructions. For now, May 2023, we recommend using CUDA 11.7 and flash-attn<1.0.5 due to various issues reported about installing new versions of flash-attn.
Pretrained scGPT Model Zoo ¶
Here is the list of pretrained models. Please find the links for downloading the checkpoint folders. We recommend using the whole-human model for most applications by default. If your fine-tuning dataset shares similar cell type context with the training data of the organ-specific models, these models can usually demonstrate competitive performance as well.
| Model name | Description | Download |
|---|---|---|
| whole-human (recommended) | Pretrained on 33 million normal human cells. | link |
| brain | Pretrained on 13.2 million brain cells. | link |
| blood | Pretrained on 10.3 million blood and bone marrow cells. | link |
| heart | Pretrained on 1.8 million heart cells | link |
| lung | Pretrained on 2.1 million lung cells | link |
| kidney | Pretrained on 814 thousand kidney cells | link |
| pan-cancer | Pretrained on 5.7 million cells of various cancer types | link |
Fine-tune scGPT for scRNA-seq integration ¶
Please see our example code in examples/finetune_integration.py. By default, the script assumes the scGPT checkpoint folder stored in the examples/save directory.
To-do-list ¶
- Upload the pretrained model checkpoint
- Publish to pypi
- Provide the pretraining code with generative attention masking
- Finetuning examples for multi-omics integration, cell type annotation, perturbation prediction, cell generation
- Example code for Gene Regulatory Network analysis
- Documentation website with readthedocs
- Bump up to pytorch 2.0
- New pretraining on larger datasets
- Reference mapping example
- Publish to huggingface model hub
Contributing ¶
We greatly welcome contributions to scGPT. Please submit a pull request if you have any ideas or bug fixes. We also welcome any issues you encounter while using scGPT.
Acknowledgements ¶
We sincerely thank the authors of following open-source projects:
Citing scGPT ¶
@article{cui2023scGPT,
title={scGPT: Towards Building a Foundation Model for Single-Cell Multi-omics Using Generative AI},
author={Cui, Haotian and Wang, Chloe and Maan, Hassaan and Pang, Kuan and Luo, Fengning and Wang, Bo},
journal={bioRxiv},
year={2023},
publisher={Cold Spring Harbor Laboratory}
}
Related notebook
BioTuring
Computational methods that model how the gene expression of a cell is influenced by interacting cells are lacking. We present NicheNet, a method that predicts ligand–target links between interacting cells by combining their expression data with prior knowledge of signaling and gene regulatory networks. We applied NicheNet to the tumor and immune cell microenvironment data and demonstrated that NicheNet can infer active ligands and their gene regulatory effects on interacting cells.
BioTuring
Cell–cell communication mediated by ligand–receptor complexes is critical to coordinating diverse biological processes, such as development, differentiation and inflammation. To investigate how the context-dependent crosstalk of different cell types enables physiological processes to proceed, we developed CellPhoneDB, a novel repository of ligands, receptors and their interactions. In contrast to other repositories, our database takes into account the subunit architecture of both ligands and receptors, representing heteromeric complexes accurately. We integrated our resource with a statistical framework that predicts enriched cellular interactions between two cell types from single-cell transcriptomics data. Here, we outline the structure and content of our repository, provide procedures for inferring cell–cell communication networks from single-cell RNA sequencing data and present a practical step-by-step guide to help implement the protocol. CellPhoneDB v.2.0 is an updated version of our resource that incorporates additional functionalities to enable users to introduce new interacting molecules and reduces the time and resources needed to interrogate large datasets. CellPhoneDB v.2.0 is publicly available, both as code and as a user-friendly web interface; it can be used by both experts and researchers with little experience in computational genomics. In our protocol, we demonstrate how to evaluate meaningful biological interactions with CellPhoneDB v.2.0 using published datasets. This protocol typically takes ~2 h to complete, from installation to statistical analysis and visualization, for a dataset of ~10 GB, 10,000 cells and 19 cell types, and using five threads.
BioTuring
Required GPU
Geneformer is a foundation transformer model pretrained on a large-scale corpus of ~30 million single cell transcriptomes to enable context-aware predictions in settings with limited data in network biology. Here, we will demonstrate a basic workflow to work with ***Geneformer*** models. These notebooks include the instruction to: 1. Prepare input datasets 2. Finetune Geneformer model to perform specific task 3. Using finetuning models for cell classification and gene classification application