Reference-Guided De Novo Retrieval of 3D Chromatin Organizational Features with Representation Learning

Chromatin Fetch (C-Fetch) retrieves 3D chromatin contact-map features from example queries, including Polycomb-guided enrichment that exceeds loop-guided assignments at 5 kb.

• Why it matters: Existing callers mainly detect predefined architectural patterns, leaving limited options when users can supply example instances but lack a dedicated detector for a feature of interest.
• What they did: C-Fetch ranks and assigns contact-map patches at candidate coordinates using a learned representation from weakly KL-regularized 3D autoencoding of structurally related, non-temporal Hi-C and Micro-C patches.
• The result: Polycomb-guided retrieval yields contacts enriched for PRC2-associated signals, exceeding loop-guided assignments at 5 kb; it also supports visual pattern queries, improves focal enrichment versus complete Mustache proposal sets across four held-out maps, and generalizes across Region Capture Micro-C and HiChIP with retained replicate agreement and reduced sequencing depth.

AI summary of the abstract · may contain errors · not medical advice · read the paper.