Bulk Hi-C¶

A bulk Hi-C map averages millions of nuclei. U-Chrom reads it (.cool, .mcool, .hic, also over HTTP), links it to a ChromData as a contact map, and turns it into 3-D in three ways: one consensus structure (MDS), a population of structures whose contacts together reproduce the map (IGM), or a model that also honours the distances of chromatin tracing in the same cell type (GEM-FISH). Each returns a ChromData, so the result is compared with imaging, measured and viewed like any traced data.

Step	API
read a map: `.cool` / `.mcool`, `.hic` (also over HTTP range requests)	`uchrom.io.load_cool`, `load_hic`, `load_hic_genome`
link a map to a `ChromData` (bulk or pseudo-bulk per cell type)	`cd.link_cool`, `cd.linked_cool_path()`
consensus structure: MDS / SMACOF (PyTorch; partitioned, inter-chromosomal)	`uchrom.recon.bulk.mds.run_mds`, `python -m uchrom.recon.bulk.mds`
population of structures: IGM (native engine)	`uchrom.recon.bulk.deconv.preprocess_hic`, `deconvolve(..., method="igm")`
Hi-C + chromatin tracing: GEM-FISH (PyTorch)	`uchrom.recon.fish.reconstruct_gem_fish`, `GEMFISHParams`
TADs from a contact map (directionality index)	`uc.tl.call_tads(cd, method="di", contacts=key)` (structure calling)

IGM runs on the native engine, the package uchrom-recon (Rust; multi-core CPU and wgpu GPU): pip install ./packages/uchrom-recon.

Tutorials¶

Guides¶

Bulk Hi-C: methods

API: uchrom.recon, GEM-FISH.