Machine Learning Unravels Inherent Structural Patterns in Escherichia Coli Hi-C Matrices and Predicts DNA Dynamics

Abstract

The large dimension of the Hi-C-derived chromosomal contact map, even for a bacterial cell, presents challenges in extracting meaningful information related to its complex organization. Here we first demonstrate that a machine-learnt (ML) low-dimensional embedding of a recently reported Hi-C interaction map of archetypal bacteria E. Coli can decode crucial underlying structural pattern. In particular, a three-dimensional latent space representation of (928x928) dimensional Hi-C map, derived from an unsupervised artificial neural network, automatically detects a set of spatially distinct domains that show close correspondences with six macro-domains (MDs) that were earlier proposed across E. Coli genome via recombination assay-based experiments. Subsequently, we develop a supervised random-forest regression model by machine-learning intricate relationship between large array of Hi-C-derived chromosomal contact probabilities and diffusive dynamics of each individual chromosomal gene. The resultant ML model dictates that a minimal subset of important chromosomal contact pairs (only 30 %) out of full Hi-C map is sufficient for optimal reconstruction of the heterogenous, coordinate-dependent sub-diffusive motions of chromosomal loci. Specifically the Ori MD was predicted to exhibit most substantial contribution in chromosomal dynamics among all MDs. Finally, the ML models, trained on wild-type E. Coli was tested for its predictive capabilities on mutant bacterial strains, shedding light on the structural and dynamic nuances of {Delta}MatP30MM and {Delta}MukBEF22MM chromosomes. Overall our results illuminate the power of ML techniques in unraveling the complex relationship between structure and dynamics of bacterial chromosomal loci, promising meaningful connections between our ML-derived insights and real-world biological phenomena.