From Protein Folds to Chromatin: Topological Data Analysis and Deep Learning at Multiple Scales

Across biological scales—from atomic-level protein folds to the three-dimensional architecture of chromatin—structural organization governs molecular function and regulation. We present two complementary computational frameworks that integrate geometric, topological, and spatial modeling with deep learning to elucidate these structure–function relationships. CASTpFold extends the Computed Atlas of Surface Topography of Proteins by combining computational geometry, topological data analysis, and AI-based structure prediction to identify and quantify surface pockets, internal cavities, and cross channels across more than 183 million experimentally determined and AlphaFold2-predicted structures. It further provides functional pocket annotations with Gene Ontology and Enzyme Commission terms and enables pocket similarity search for surface and interface comparison through the CASTpFold web server. At the chromatin scale, CHROME introduces a chromatin-structure–guided graph attention framework built on a self-avoiding polymer ensemble null model that identifies physically specific, non-random Hi-C contacts. By integrating DNA sequence, chromatin accessibility, or pre-trained embeddings, CHROME predicts transcription factor binding and histone modification landscapes with high accuracy, generalizes across cell lines, and reveals how distal chromatin interactions influence local regulatory activity. Together, we bring a unified multi-scale approach that couples geometric, topological, and spatial representations with deep learning to uncover the structural principles linking protein form and genome architecture to biological regulation.

Author dc:creator

• Bowei Ye (17695709)

• Bioinformatics

Statement dc:rights

• In Copyright

DOI dc:identifier

https://doi.org/10.25417/uic.32991884.v1

OAI identifier oai:identifier

oai:figshare.com:article/32991884