Article

Decoding Conserved and Tissue-Specific Cellular Stress Programs from Single-Cell Foundation Models Using Sparse Autoencoders

2026-04-28

Abstract excerpt

Single-cell foundation models (scFMs) learn rich, generalizable representations of cellular states across tissues and conditions, but their latent spaces remain notoriously difficult to interpret. Here, we apply sparse autoencoders (SAEs) to disentangle scFM embeddings into human-readable, monosemantic features corresponding to distinct cellular stress programs. Using a publicly available multi-organ single-cell a...

Topics

Open a Topic to create a Post that cites this publication.

Identifiers and source

Literature Corpus work: cb331e4f-e9d1-5575-8354-659e5e4f1fc4
DOI: 10.14293/pr2199.003483.v1

Open publication

Related research

Semantic proximity does not establish scientific evidence.

Click a neighbor to travelStep 1 · 12 closest

Decoding Conserved and Tissue-Specific Cellular Stress Programs from Single-Cell Foundation Models Using Sparse AutoencodersDOI 10.14293/pr2199.003483.v1Read

Inflammation-Linked Aging Signals in Frozen Single-Cell Foundation Models: Donor-Aware Detection and Robustness TestingSemantic proximity Read

Sparse autoencoders reveal organized biological knowledge but minimal regulatory logic in single-cell foundation models: a comparative atlas of Geneformer and scGPTSemantic proximity Read

Regulatory Logic in Larger Single-Cell Foundation Models: A Sparse Autoencoder Comparison of scFoundation and GeneformerSemantic proximity Read

TRAILBLAZER: generative multicellular perturbation model of biologySemantic proximity Read

Scalable Non-negative Matrix Factorization of the Human Cell Census Reveals Interpretable Transcriptional ProgramsSemantic proximity Read

Large Scale Foundation Model on Single-cell TranscriptomicsSemantic proximity Read

Multiscale transcriptomic organization of the human brain with DigitalBrainSemantic proximity Read

Low Transcriptional Complexity Cells Represent a Conserved, Aging-Relevant Maintenance State Overlooked by Single-Cell TranscriptomicsSemantic proximity Read

A modular architecture of human cellular aging across organs and diseasesSemantic proximity Read

scAgeClock: a single-cell transcriptome based human aging clock model using gated multi-head attention neural networksSemantic proximity Read

Decomposing multi-scale dynamic regulation from single-cell multiomics with scMagnifySemantic proximity Read