/scMulan

Repository for paper scMulan: a multitask generative pre-trained language model for single-cell analysis.

Primary LanguageJupyter Notebook

scMulan

Welcome to the repository for scMulan_v1, featuring our upcoming work: "scMulan: A Multitask Generative Pre-trained Language Model for Single-Cell Analysis."

Introducing scMulan

scMulan is a groundbreaking foundation model for the analysis of single-cell transcriptomics.
Features:

  • Zero-shot Cell Type Annotation: Effortlessly classify cell types without prior training or reference mapping.
  • Zero-shot Batch Integration: Seamlessly integrate data from diverse batches.
  • Conditional Cell Generation: Support simulation of in-silico perturbations.

Installation

conda create -n scMulan python==3.10
conda activate scMulan
pip install -r requirements.txt

Quick start

download the ckpt file and put it under ./ckpt/
Prepare your test adata file, and start using scMulan

Tutorials

We provided a tutorial of using scMulan for cell type annotation (see tutorial). Currently, scMulan supports zero-shot annotation of human cell types in seven organs including Heart, Lung, Liver, Bone marrow, Blood, Brain, and Thymus.

It could also be used to get cell embeddings for batch integration (see tutorial). You can easily use your adata and get analysis from scMulan.

scMulan now supports inference on npu.

Acknowledgements