scGPT: toward building a foundation model for single-cell multi-omics using generative AI

TL;DR

scGPT is a generative transformer-based foundation model pretrained on single-cell transcriptome data, enabling transfer learning for diverse single-cell analysis tasks.

Problem

Single-cell data is high-dimensional and noisy, with batch effects varying across experiments, making integrated analysis challenging. Additionally, tasks like perturbation response prediction and gene network inference suffer from data scarcity.

Approach

Drawing an analogy between words in text and genes in cells, the generative pretrained transformer (GPT) architecture is applied to single-cell RNA sequencing data. The model is pretrained on over 33 million cell profiles and then fine-tuned for specific downstream tasks.

Results & Contribution

scGPT achieves superior or competitive performance compared to existing methods in tasks such as cell type annotation, batch integration, multi-omic integration, perturbation response prediction, and gene network inference. It demonstrates the potential of foundation models in single-cell biology.