scCotag: Prior-Informed Co-Optimal Transport for Robust Single-Cell Multi-Omics Integration
Year: 2026
Type: conference-paper
Abstract: Recent advances in high-throughput single-cell technologies enable characterization of cellular states across distinct omic layers, yielding complementary insights into complex biological systems. To jointly leverage these disparate modalities, computational methods have been developed to align cells profiled in different omic layers, a task known as diagonal integration. However, existing methods often retain noisy, context-dependent biological priors as static ground truths, compromising their alignment accuracy. Moreover, they struggle to handle dataset imbalance by forcing over-alignment, thus distorting the underlying data geometry. To address these limitations, we propose scCotag, a deep generative framework leveraging prior-informed co-optimal transport (COOT) for rigorous diagonal integration of single-cell multi-omics data. scCotag iteratively adapts biological priors to observed data, yielding data-calibrated cell and feature correspondences. In addition, scCotag performs transport-guided representation learning and incorporates a novel imbalance-aware alignment regularization to mitigate geometry distortion in partially overlapping integration. Experimental results demonstrate scCotag's superior performance in alignment accuracy across three multi-omics datasets. In particular, scCotag excels at preserving geometric structure while maintaining robust alignment under simulated imbalance scenarios involving modality-unique cell types. Applying scCotag to unpaired multiome data from postmortem brains of Alzheimer's disease (AD) and control patients further identifies dysregulated cis-regulatory interactions and regulons, including the activation of ATF6, SPI1, and increased activity of ZNF652, in AD pathogenesis.
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7196955674