Hypothesis-driven probabilistic modelling enables a principled perspective of genomic compartments

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Hi-C methods offer insights into genome organization but interpreting interaction maps is challenging.
• Genomic compartments, characterized by checkered Hi-C patterns, are thought to reflect active and inactive chromatin states.
• Understanding the mechanistic basis of these compartments is crucial for deciphering genome architecture.

Purpose of the Study:

• To develop a generative probabilistic model for genomic compartments.
• To robustly interpret Hi-C interaction frequency maps, even with sparse data.
• To investigate the nature and molecular underpinnings of genomic compartments.

Main Methods:

• Derivation of a probabilistic model (deGeco) based on mechanistic assumptions.
• Testing deGeco's ability to explain and infer Hi-C interaction maps from sparse data.
• Analysis of model parameters to test hypotheses about chromatin state interactions.
• Classification of chromatin states using histone marks.

Main Results:

• deGeco accurately explains Hi-C interaction maps and infers interaction probabilities from sparse data without parameter training.
• Evidence for multiple self-interacting chromatin states with differing affinities was found.
• Chromatin state interaction rules vary significantly within and between chromosomes.
• A classifier using histone marks predicts underlying chromatin states with 87% accuracy.
• Mixed-state loci were observed, with state mixing primarily occurring at the cellular level.

Conclusions:

• deGeco provides a robust framework for modeling and interpreting genomic compartments from Hi-C data.
• The study provides evidence for complex, multi-state genome organization with distinct interaction dynamics.
• Histone modifications are strong predictors of underlying chromatin states, offering molecular insights into genome compartmentalization.