TSProm: deep learning framework to predict tissue-specific regulatory logic
NAR Genomics and Bioinformatics · 27.03.2026 · 8 (2)
DOI: 10.1093/nargab/lqag050 ↗Kurz gefasst
Fachartikel in der Zeitschrift NAR Genomics and Bioinformatics, online erschienen am 03.06.2026. Verfasst von Pallavi Surana und 6 weiteren Autor:innen. Beteiligte Einrichtung: Department of Biomedical Informatics, Stony Brook University, NY 11794, United States. Hauptthemen laut MEDLINE: Deep Learning, Promoter Regions, Genetic, Gene Expression Regulation. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Institutes of Health, National Library of Medicine.
Abstract (Originalsprache)
Characterizing tissue-specific (TSp) gene expression is crucial for understanding development and disease; however, traditional expression-based methods often overlook the latent “regulatory grammar” embedded in non-coding DNA, particularly across distal promoter regions. Here, we introduce TSProm, a framework that adapts a DNA foundation model (DNABERT2) to decode the regulatory logic of TSp promoters at the isoform level. Our contributions are two-fold: (i) a comparative design that trains two specialized models: Model A for general promoter biology and Model B for tissue-specific regulation enabling precise isolation of sequence motifs surrounding the transcription start site that uniquely define tissue identity; (ii) we develop an explainable AI module that integrates attention-based motif discovery with model-agnostic SHAP analysis to yield cross-validated interpretations of learned features. Applying TSProm to human brain, liver, and testis promoters, we identified clinically relevant transcription factors (TFs) in the brain, including SP1, MYC, and HES6, whose associations with gliomas and neuroblastomas highlight clinical relevance. Moreover, our results highlight C2H2 zinc finger proteins as a dominant family shaping the global landscape of TSp gene regulation. TSProm provides an interpretable and generalizable framework for identifying tissue-specific regulatory elements, offering powerful computational tools to investigate gene regulation in both normal and disease contexts.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Substanzen
Autor:innen und Institutionen
- Department of Biomedical Informatics, Stony Brook University, NY 11794, United States.
- Department of Computer Science, Northwestern University, IL 60208, United States.
ORCID-Profile: Pallavi Surana, Ramana V Davuluri
Förderung
- National Institutes of Health
- National Library of Medicine
Zitieren
APA
Surana, P., Dutta, P., Papineni, N., Sathian, R., Zhou, Z., Liu, H., & Davuluri, R. V. (2026). TSProm: deep learning framework to predict tissue-specific regulatory logic. NAR Genomics and Bioinformatics, 8(2). https://doi.org/10.1093/nargab/lqag050
Vancouver
Surana P, Dutta P, Papineni N, Sathian R, Zhou Z, Liu H, et al. TSProm: deep learning framework to predict tissue-specific regulatory logic. NAR Genomics and Bioinformatics. 2026;8(2). doi:10.1093/nargab/lqag050
BibTeX
@article{Surana2026,
author = {Surana, Pallavi and Dutta, Pratik and Papineni, Nimisha and Sathian, Rekha and Zhou, Zhihan and Liu, Han and Davuluri, Ramana V},
title = {TSProm: deep learning framework to predict tissue-specific regulatory logic},
journal = {NAR Genomics and Bioinformatics},
year = {2026},
volume = {8},
number = {2},
doi = {10.1093/nargab/lqag050}
}
EUVIMED