⚙
AI-generated deep summary
by claude@2026-07, 2026-07-04
· read from full text
ⓘ
The paper introduces EpiAgent, described as the first foundation model for single-cell epigenomic data, focusing on chromatin accessibility (scATAC) rather than the more commonly modeled single-cell transcriptome. It reports pretraining on a large Human-scATAC-Corpus of about 5 million cells and 35 billion tokens, using an approach that encodes accessibility patterns as “cell sentences” and applies bidirectional attention to capture cellular heterogeneity, with external embeddings to support response, integration, mapping, and perturbation prediction. The authors benchmark the model across downstream tasks including unsupervised feature extraction, supervised cell annotation, data imputation, and zero-shot cell type annotation, and they simulate cis-regulatory element knockouts for in-silico cancer analysis. The paper’s main caveat is that the approach is specific to single-cell epigenomic (chromatin accessibility) data and does not address other modalities such as single-cell gene expression in its core design. The paper does not explicitly discuss endometriosis or adenomyosis; it was included in the corpus via a keyword match in the upstream search index.
Abstract
Large-scale foundation models have recently opened new avenues for artificial general intelligence. Such a research paradigm has recently shown considerable promise in the analysis of single-cell sequencing data, while to date, efforts have centered on transcriptome. In contrast to gene expression, chromatin accessibility provides more decisive insights into cell states, shaping the chromatin regulatory landscapes that control transcription in distinct cell types. Yet, challenges also persist due to the abundance of features, high data sparsity, and the quasi-binary nature of these data. Here, we introduce EpiAgent, the first foundation model for single-cell epigenomic data, pretrained on a large-scale Human-scATAC-Corpus comprising approximately 5 million cells and 35 billion tokens. EpiAgent encodes chromatin accessibility patterns of cells as concise “cell sentences,” and employs bidirectional attention to capture cellular heterogeneity behind regulatory networks. With comprehensive benchmarks, we demonstrate that EpiAgent excels in typical downstream tasks, including unsupervised feature extraction, supervised cell annotation, and data imputation. By incorporating external embeddings, EpiAgent facilitates the prediction of cellular responses to both out-of-sample stimulated and unseen genetic perturbations, as well as reference data integration and query data mapping. By simulating the knockout of key cis -regulatory elements, EpiAgent enables in-silico treatment for cancer analysis. We further extended zero-shot capabilities of EpiAgent, allowing direct cell type annotation on newly sequenced datasets without additional training.
Full text
1,748 characters
· extracted from
oa-doi-fallback
· click to expand
Abstract
Large-scale foundation models have recently opened new avenues for artificial general intelligence. Such a research paradigm has recently shown considerable promise in the analysis of single-cell sequencing data, while to date, efforts have centered on transcriptome. In contrast to gene expression, chromatin accessibility provides more decisive insights into cell states, shaping the chromatin regulatory landscapes that control transcription in distinct cell types. Yet, challenges also persist due to the abundance of features, high data sparsity, and the quasi-binary nature of these data. Here, we introduce EpiAgent, the first foundation model for single-cell epigenomic data, pretrained on a large-scale Human-scATAC-Corpus comprising approximately 5 million cells and 35 billion tokens. EpiAgent encodes chromatin accessibility patterns of cells as concise “cell sentences,” and employs bidirectional attention to capture cellular heterogeneity behind regulatory networks. With comprehensive benchmarks, we demonstrate that EpiAgent excels in typical downstream tasks, including unsupervised feature extraction, supervised cell annotation, and data imputation. By incorporating external embeddings, EpiAgent facilitates the prediction of cellular responses to both out-of-sample stimulated and unseen genetic perturbations, as well as reference data integration and query data mapping. By simulating the knockout of key cis-regulatory elements, EpiAgent enables in-silico treatment for cancer analysis. We further extended zero-shot capabilities of EpiAgent, allowing direct cell type annotation on newly sequenced datasets without additional training.
Competing Interest Statement
The authors have declared no competing interest.
Text is read by the "Ask this paper" AI Q&A widget below.
Extraction quality varies by source — PMC NXML preserves structure
cleanly, OA-HTML may include some navigation residue, and OA-PDF can
have broken hyphenation. The publisher copy
(via DOI)
is the canonical version.