Our article is now published in Nature Methods.
We developed a large-scale pretrained model scFoundation with 100M parameters. scFoundation was based on the xTrimoGene architecture and trained on over 50 million human single-cell transcriptomics data, which contain high-throughput observations on the complex molecular features in all known types of cells. scFoundation is a large-scale model in terms of the size of trainable parameters, dimensionality of genes and the number of cells used in the pre-training. Experiments showed that scFoundation can serve as a foundation model for single-cell transcriptomics and achieve state-of-the-art performances in a diverse array of downstream tasks.
Update: Please note that the old platform was officially discontinued on April 30th, 2024. We kindly request you to migrate to our new platform at https://aigp.biomap.com/. You can follow the tutorial on the platform now. This upgrade aims to provide you with both online inference service and command-line interface(CLI) tools.
For any API issues, please get in touch with [email protected]
We now provide model pretrained weight and code with documentation of obtaining the cell embeddings and fine-tuning/integrating our model with other models. Please find the further instructions in the model folder.
This repository provides the source code necessary to use the scFoundation generated cell and gene embeddings for several downstream tasks such as gene expression enhancement, drug response prediction and perturbation prediction. The source codes for the downstream tasks are in the following repositories:
The results of SAVER, scImpute, MAGIC were obtained from the SAVER repository (https://github.com/mohuangx/SAVER-paper). The results of scFoundation were obtained by running the bash run.sh in the enhancement folder. You can find details in the enhancement/README.md.
The baseline code is from https://github.com/kimmo1019/DeepCDR
Please follow the commands in DeepCDR/prog/run.sh. The scFoundation embeddings of Bulk data are at DeepCDR/data/50M-0.1B-res_embedding.npy. You can find details in the DeepCDR/README.md.
The baseline code is from https://github.com/CompBioT/SCAD
Please follow the steps detailed in SCAD/README.md. The scFoundation embeddings of Bulk and single cell data are in the SCAD/data/split_norm/ folder.
The baseline code is from https://github.com/snap-stanford/GEARS.
The commands required for running the code can be found in GEARS/run_sh. The gene embedding of each cell is 19264*512 which is too large to be saved. We generated the gene embedding during the training process. You can find details in the GEARS/README.md.
In the genemodule directory, you can find the code for inferring gene modules from gene context embeddings.
The mapping directory contains the demo usage code and scripts to reproduce figures related to the cell mapping task.
You can find the code to reproduce results for the cell type annotation task in the annotation folder.
We provide the code for downloading and processing the data used for pre-training. The code and demo usage are in the preprocessing folder.
| Task and Functions | Description | Code path | Data path |
|---|---|---|---|
| Ablation | Ablation study on different model and loss settings | ablation folder | Figshare: data_ablation.zip |
| Annotation | Cell type annotation task on Pancreatic and PBMC data | annotation folder | GitHub (embeddings): annotation/annotation_data.zip & Figshare (raw data): cell_type_rawdata.zip |
| API | Instruction for using API | apiexample folder | GitHub: apiexample/data/ folder |
| DeepCDR | Cancer Drug Response (IC50) prediction | DeepCDR folder | GitHub: DeepCDR/data/ folder |
| Read Depth Enhancement | Enhancing cells' read depth for clustering | enhancement folder | GitHub: enhancement folder |
| GEARS | perturbation prediction | GEARS folder | GitHub (demo data): GEARS/demo/data/ folder & Figshare (Experiment data): all h5ad files |
| Gene Module | Inferring gene modules and regulation networks | genemodule folder | Figshare: data_genemodule.zip |
| Data Mapping | Mapping organoid data into in vivo data | mapping folder | Figshare: data_mapping.zip |
| Model Code | Using pretrained model for embedding inference/ for integrating/finetuning with other models | model folder | Figshare: model_example.zip |
| Pretraining Data Processing | single cell RNA-seq data collection workflow | preprocessing folder | DataSupplement1.xlsx and DataSupplement2.xlsx |
| SCAD | single cell level drug sensitivity prediction | SCAD folder | GitHub (embeddings): SCAD/data/split_norm Figshare (raw exp. Data): data_SCAD_split_norm.zip |
This source code repository is licensed under the Apache-2.0 License; provided, however, that notwithstanding the foregoing, any use or exploitation of any Model Weights is subject to the Model License.
Use of the third-party software, libraries or code referred to in the Acknowledgements section may be governed by separate terms and conditions or license provisions.
Your use of the third-party software, libraries or code is subject to any such terms and you should check that you can comply with any applicable restrictions or terms and conditions before use.
scFoundation uses and/or references the following separate libraries and packages (ordered alphabetically):
Thanks for all their contributors and maintainers!