Dynamix identifies accurate chromatin stripes by Bayesian Normal Dynamic Linear Model on Hi-C signal probabilities
Year: 2026
Type: preprint
Abstract: Background Chromatin stripes, characterized by interactions between a localized genomic region (the stripe anchor) and a broader, continuous genomic segment (the stripe length range), are topological features visible in Hi-C interaction matrices. Existing chromatin stripe-calling algorithms often mislocalize stripes into regions of low interaction probability by operating on Hi-C normalized matrices without proper bias removal, highlighting the need to introduce more precise tool. Results Here, I present Dynamix, a software designed to call chromatin stripes through a two-step statistical framework. First, it computes the signal probabilities of genomic interactions using a Zero-Inflated Negative Binomial and Negative Binomial mixture expectation-maximization algorithm, while adjusting for genomic distance, GC ratio, and mappability biases. Second, utilizing this signal probability matrix, it identifies stripe anchors and length ranges via a Bayesian Normal Dynamic Linear Model. Dynamix detects the largest number of stripes, and these stripes accurately capture enhancer-promoter interactions across both their anchors and length ranges, successfully uncovering long-range enhancer-promoter interactions even beyond topologically associating domain boundaries, compared to all evaluated tools. Conclusions Dynamix provides a precise tool for identifying chromatin stripes, facilitating a deeper investigation of dynamic loop formation and enhancer-promoter interactions in cellular environments.
Source: Research Square
Author: Soomin Han
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225740908
