PROBC

PROBC performs probabilistic decomposition of chromatin marks to jointly analyze epigenomic and transcriptomic contributions to three-dimensional genome interactions inferred from Hi-C and Micro-C data.

Key Features:

• Integration of Chromatin Marks: Integrates histone modifications H3K27ac, H3K9me3, H3K4me3, H3K4me1 and CTCF binding sites to attribute contributions of chromatin marks to 3D interactions.
• Convex Likelihood Optimization: Employs a probabilistic model based on convex likelihood optimization to decompose interaction matrices according to known chromatin marks.
• Modeling of Present and Absent Interactions: Simultaneously models both observed and unobserved (existing and non-existing) Hi-C and Micro-C interactions.
• Predictive Power: Predicts Hi-C and Micro-C interactions with reported superior performance relative to existing methods, leveraging histone modifications over transcription factor binding sites for explanatory power.
• Cross-Species Prediction: Enables cross-species prediction of nucleosome-resolution Micro-C interactions (e.g., human ES cells from mouse ES training) with reported AUC > 0.75.
• Optimal Decomposition: Performs optimal decomposition at genome and chromosome levels to identify subsets of histone modifications and transcription factor binding sites predictive of topologically associating domains (TADs).
• Application in Limited Data Scenarios: Predicts 3D interactions from chromatin marks alone when Hi-C data are limited or unavailable for a species.

Scientific Applications:

• Interaction Prediction: Predicts Hi-C and Micro-C contact probabilities from epigenomic marks and CTCF to infer 3D genome architecture.
• Cross-Species Inference: Transfers interaction models between species, demonstrated by predicting human nucleosome-resolution Micro-C interactions using mouse ES cell data.
• TAD-Associated Mark Identification: Identifies histone modification and transcription factor binding site subsets that are predictive of TADs across cell types and species.
• Inference with Limited Hi-C: Infers genome-wide interaction patterns in species or conditions lacking Hi-C by using available chromatin modification profiles.

Methodology:

PROBC implements a probabilistic decomposition framework using convex likelihood optimization to decompose Hi-C and Micro-C interaction matrices based on input chromatin marks, explicitly modeling both presence and absence of interactions.