Persistent Homology on Hi-C Contact Matrices

Topological data analysis pipeline for chromatin loop and TAD detection in Hi-C contact matrices, extending TDA Engine v1.0 into 3D genomic architecture.

Status

Early stage. Core persistent homology computation on public Hi-C datasets is in progress. Built to scale via parallel fan-out across chromosomes, resolutions, and replicates (Modal compute).

Goal

Compute persistent homology on Hi-C contact matrices to identify chromatin loops and TADs as topologically stable features, rather than local statistical peak calls. Cross-validate feature stability across resolution (5kb-25kb) and replicate perturbation. Correlate topologically stable loop anchors against ESM-2 sequence embeddings at CTCF/cohesin binding sites.

Approach

1. Ingest .hic contact matrices via the Juicebox/straw ecosystem
2. Distance-transform contact matrices, compute persistence diagrams
3. Apply bottleneck/Wasserstein distance diagnostics across resolutions and replicates to separate stable features from noise
4. Benchmark against HiCCUPS/Arrowhead loop and TAD calls

Related work

• tda-foundation-models extending topological probing of ESM-2 latent geometry, feeding the sequence correlation step here
• TDA Engine v1.0, core infrastructure this pipeline extends

Data

Public Hi-C datasets (4DN, ENCODE) at chromosome-level resolution. Not committed to this repo; download scripts in src/.