Abstract

Background

Accurately identifying functionally distinct tumor cell subpopulations remains a critical challenge in cancer research. While single-cell epigenomics assays provide powerful insights into tumor heterogeneity beyond gene expression, computational limitations have hindered their application.

Methods

We introduce Multimodal-based Analysis of scATAC-Seq data (MAAS), a method that integrates chromatin accessibility, copy number variations (CNVs), and single-nucleotide variants (SNVs) to identify functional tumor cell subpopulations. MAAS employs a self-expressive multimodal matrix factorization approach with rigorous coverage normalization and data denoising. We applied MAAS to simulated datasets and multiple real-world tumor scATAC-seq datasets, including pediatric ependymoma, B-cell lymphoma, and glioblastoma, and benchmarked its performance against existing integration methods. Functional relevance of subpopulation-specific genes was experimentally validated using gene knockdown and overexpression assays. Furthermore, we constructed subpopulation-specific gene regulatory networks and developed a prognostic signature from the key regulatory genes.

Results

MAAS demonstrated superior accuracy in detecting clinically relevant subpopulations, particularly in tumors with limited CNV heterogeneity, such as pediatric ependymoma and B-cell lymphoma. In glioblastoma, MAAS uncovered a previously unrecognized subpopulation with temozolomide resistance and further experimentally validated the effects of its signature genes through gene knockdown and overexpression. The MAAS-derived prognostic signature, MAASig, outperformed traditional clinicopathologic features across multiple cancer types when applied to independent validation cohorts.

Conclusions

By integrating multimodal information from scATAC-seq data, MAAS provides the robust identification of functionally distinct tumor cell subpopulations, facilitating the discovery of potential therapeutic targets.

Title

Multimodal-based analysis of single-cell ATAC-seq data enables highly accurate delineation of clinically relevant tumor cell subpopulations

Authors

Kewei Xiong, Wei Wang, Ruofan Ding, Dinglin Luo, Yangmei Qin, Xudong Zou, Jiguang Wang, Chen Yu & Lei Li

Journal Information

Genome Medicine (2026)

View at publisher↗