Authors
Zhenglong Cheng, Jiao Zhang, Shixiong Zhang
Published in
Bioinformatics (Oxford, England). Sep 18, 2026. Epub Sep 18, 2026.
Abstract
Single-cell RNA sequencing and single-cell ATAC sequencing provide complementary views of transcriptional output and chromatin regulatory potential, but integrating unpaired profiles remains challenging because the modalities differ in feature space, sparsity and noise. Existing approaches often frame integration as distribution matching, which can over-align biologically distinct, condition-specific or modality-specific cell states. We present scHPGT, a single-cell Heterogeneous Prior-Guided Transformer for regulatory-prior-guided integration of unpaired RNA and chromatin accessibility profiles. scHPGT uses modality-specific encoders to model RNA and ATAC signals, a prior-guided cross-modal Transformer to constrain gene-peak attention using regulatory links, and a domain-adversarial objective to reduce modality-specific discrepancies in a shared latent space.
Across PBMC3k, mouse spleen, CITE-seq/ASAP-seq PBMC and PBMC10k benchmarks, scHPGT improves clustering agreement, label transfer and biological structure preservation while maintaining effective modality alignment. In partial-overlap and condition-shift settings, scHPGT aligns shared populations without forcing unmatched or condition-specific states into inappropriate correspondence. Attention-derived links recover regulatory relationships, highlight marker-gene regulatory regions, recover transcription factor programs and produce regulatory activity profiles consistent with cell-type-specific transcriptional programs.
Code and datasets are released at https://github.com/zlCreator/scHPGT.
Supplementary data are available at Bioinformatics online.
PMID:
42758136
Bibliographic data and abstract were imported from PubMed on 18 Sep 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 30
- Comments 0