scLong: a billion-parameter foundation model for capturing long-range gene context in single-cell transcriptomics

2026-02-05 · Source: Machine learning : nature.com subject feeds · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Computational Biology · Depth: Expert, long

Summary

scLong is a new billion-parameter foundation model designed for single-cell RNA sequencing (scRNA-seq) data analysis, pretrained on 48 million cells. Published in Nature Communications on February 5, 2026, it addresses limitations of existing models by performing self-attention across all 28,000 genes in the human genome. This capability allows scLong to capture long-range dependencies, including those involving lowly expressed genes often overlooked by other models, which are critical for cellular processes. Additionally, scLong integrates external gene knowledge from the Gene Ontology using a graph convolutional network to enhance its understanding of gene functions. Extensive evaluations demonstrate that scLong outperforms both current state-of-the-art scRNA-seq foundation models and task-specific models across various applications, such as predicting transcriptional responses to perturbations, forecasting cancer drug responses, and inferring gene regulatory networks.

Key takeaway

For AI Researchers and Computational Biologists working with single-cell transcriptomics, scLong offers a powerful new tool to overcome limitations of previous foundation models. Its ability to analyze all 28,000 human genes and integrate external biological knowledge means your research can uncover more comprehensive insights into cellular processes, especially those driven by lowly expressed genes. Consider adopting scLong for more accurate predictions in drug response and gene regulatory network inference.

Key insights

scLong is a billion-parameter foundation model for scRNA-seq, capturing long-range gene dependencies and integrating external gene knowledge.

Principles

• Model all genes, including lowly expressed ones.
• Integrate external biological knowledge for context.

Method

scLong employs self-attention across 28,000 human genes to capture long-range dependencies and integrates Gene Ontology knowledge via a graph convolutional network for enriched contextual understanding.

In practice

• Predict transcriptional responses to perturbations.
• Forecast cancer drug responses.
• Infer gene regulatory networks.

Topics

• scLong
• Single-cell Transcriptomics
• Foundation Models
• Gene Regulatory Networks
• Graph Convolutional Networks

Code references

• snap-stanford/GEARS
• njpipeorgan/L1000-bayesian
• kimmo1019/DeepCDR
• Sinwang404/DeepDDS
• HantaoShu/DeepSEM

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• scBench-Long: Verifiable Benchmarking of Long-Horizon Single-Cell Biology — scBench-Long evaluates AI agents' ability to derive complex scientific conclusions from raw single-cell data, revealing current limitations.
• scGTN: Deep Siamese Graph Transformer Network for Single-cell RNA Sequencing Clustering — scGTN integrates gene expression and intercellular structure via a Siamese graph transformer for robust scRNA-seq clustering.
• Teaching an old dog new cells — Bulk-trained sequence models can be efficiently adapted for single-cell gene regulation and variant effect prediction.
• Why the Human Genome’s Tangled Physicality May Confound AI — The human genome's dynamic, multi-layered regulation defies simple linear models, challenging AI's current sequence-based predictive approaches.
• Discovery of candidate therapeutic targets with Geneformer — Geneformer leverages a transformer model and transfer learning for gene network analysis and therapeutic target discovery with limited data.
• SEAGAN: domain-Specific and Edge-Aware Graph Attention Network for Dynamic Plant Processes — SEAGAN, an edge-aware graph attention network, accurately identifies biochemical limitation states in plant A-Ci curves.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.