--- id: ai-for-genomics title: "AI for Genomics: Single-Cell Foundation Models and RNA Biology" schema_type: article category: ai language: en confidence: high last_verified: "2026-05-24" created_date: "2026-05-24" generation_method: ai_assisted ai_models: - claude-4.5-sonnet derived_from_human_seed: true conflict_of_interest: none_declared is_live_document: false data_period: static completeness: 0.85 atomic_facts: - id: af-ai-for-genomics-1 statement: >- Nature Experimental & Molecular Medicine (2025) published a comprehensive review of single-cell foundation models — large pretrained models (scGPT, scBERT, Geneformer) trained on massive single-cell RNA-sequencing datasets from millions of cells across diverse tissues and species — showing that these models capture universal gene expression patterns and enable zero-shot cell type annotation, perturbation response prediction, and cross-species transfer learning. source_title: Nature Experimental & Molecular Medicine (2025) — Single-cell foundation models review — doi:10.1038/s12276-025-01547-5 source_url: https://www.nature.com/articles/s12276-025-01547-5 confidence: high - id: af-ai-for-genomics-2 statement: >- T&F RNA Biology (March 2026) published a landmark review documenting AI foundation models for RNA biology — showing that models pretrained on RNA sequences predict secondary structure, RNA-protein interactions, and alternative splicing with accuracy rivaling experimental methods, while generating functional RNA sequences de novo. source_title: Taylor & Francis RNA Biology (2026) — AI foundation models for RNA biology — doi:10.1080/15476286.2026.2650517 source_url: https://www.tandfonline.com/doi/full/10.1080/15476286.2026.2650517 confidence: high primary_sources: - id: ps-ai-for-genomics-1 title: "Single-cell foundation models: bringing artificial intelligence to single-cell biology" type: academic_paper year: 2025 institution: Nature Experimental & Molecular Medicine doi: 10.1038/s12276-025-01547-5 url: https://www.nature.com/articles/s12276-025-01547-5 - id: ps-ai-for-genomics-2 title: AI foundation models for RNA biology type: academic_paper year: 2026 institution: Taylor & Francis RNA Biology doi: 10.1080/15476286.2026.2650517 url: https://www.tandfonline.com/doi/full/10.1080/15476286.2026.2650517 known_gaps: - Integrating multi-omics data (proteomics, metabolomics) into unified genomic foundation models - Clinical translation of genomic AI predictions into diagnostic tools disputed_statements: [] secondary_sources: - title: "AlphaFold 3: Accurate Structure Prediction of Biomolecular Interactions (DeepMind)" type: journal_article year: 2024 authors: - Abramson, Josh - Adler, Jonas - Dunger, Jack - et al. institution: Google DeepMind / Nature url: https://www.nature.com/articles/s41586-024-07487-w - title: Effective Gene Expression Prediction from Sequence by Integrating Long-Range Interactions (Enformer) type: journal_article year: 2021 authors: - Avsec, Žiga - Agarwal, Vikram - Visentin, Daniel - et al. institution: Google DeepMind / Nature Methods url: https://doi.org/10.1038/s41592-021-01252-x - title: "Deep Learning in Genomics: A Comprehensive Survey and Outlook" type: survey_paper year: 2024 authors: - multiple institution: Nature Reviews Genetics url: https://doi.org/10.1038/s41576-024-00712-6 - title: "Large Language Models in Genomics: From DNA Language Modeling to Protein Design" type: survey_paper year: 2025 authors: - multiple institution: Genome Biology (BioMed Central) url: https://doi.org/10.1186/s13059-025-03456-7 updated: "2026-05-24" --- ## TL;DR AI foundation models trained on massive genomic datasets — from single-cell RNA sequencing of millions of cells to full RNA transcriptomes — are revolutionizing genomics. These models learn universal biological representations, enabling zero-shot cell typing, perturbation prediction, and de novo RNA design that were previously impossible without labor-intensive experiments. ## Core Explanation Single-cell genomics generates high-dimensional data: gene expression profiles for individual cells (20,000+ genes × millions of cells). Traditional analysis clusters cells by expression similarity and manually annotates cell types. AI foundation models (scGPT, scBERT, Geneformer, scFoundation) pretrain on massive scRNA-seq compendia using transformer architectures with gene-as-token representations. Key capabilities: (1) Cell type annotation — zero-shot classification of novel cell populations without labeled training data; (2) Perturbation prediction — predict how gene expression changes when specific genes are knocked out or drugs are applied; (3) Cross-species transfer — apply knowledge from mouse to human cell atlases; (4) Gene program discovery — identify co-regulated gene modules across conditions. ## Detailed Analysis scGPT (2023, Toronto) uses a generative pretraining objective — masking gene expression values and predicting them from context, analogous to BERT's masked language modeling. scBERT adapts the BERT architecture with gene-specific tokenization. Geneformer (Theodoris et al., 2023, Nature) demonstrated that pretraining on 30M single-cell transcriptomes enables context-aware gene network predictions. ESM-2 and ESM-3 (Meta/EvolutionaryScale, 2023-2024) extended the paradigm to protein sequences, generating functional proteins at scale. RNA foundation models (2026) integrate sequence, secondary structure, and experimental binding data, predicting RNA-small molecule interactions critical for RNA-targeted therapeutics. Key infrastructure: the Human Cell Atlas (HCA), Tabula Sapiens, and CELLxGENE provide the training data; BioLLM (2025, ScienceDirect) standardizes single-cell analysis pipelines. Critical challenge: batch effects across labs and sequencing platforms degrade model generalization; specialized normalization and harmonization methods remain active research areas. ## Further Reading - Geneformer: Transfer Learning for Gene Networks (Theodoris et al., Nature 2023) - scGPT GitHub: bowang-lab/scGPT - HCA: Human Cell Atlas Data Portal