PhiMRF

PhiMRF models spatial dependencies in count-based gene expression data using a Poisson Hierarchical Markov Random Field (PHMRF) framework to quantify how 3D genome architecture influences gene expression.

Key Features:

• Integration of Multi-Omics Data: Integrates RNA-seq data with HiC outputs to link linear and 3D genomic contexts to gene expression counts.
• Poisson Hierarchical Markov Random Field (PHMRF): Models count data using a hierarchical Poisson MRF framework to represent spatial autocorrelation among loci.
• Bayesian Inference Framework: Employs Bayesian estimation to infer spatial dependencies among protein-coding genes.
• Spatial Interaction Estimate (SIE): Computes the SIE metric to quantify the strength of spatial dependencies, including intra- and inter-chromosomal interactions.
• Detection of Chromatin Structures: Identifies positive intra-chromosomal spatial dependencies associated with chromatin loops and Topologically Associating Domains (TADs), and assesses TAD boundary insulation effects.
• Inter-Chromosomal Correlations: Estimates high inter-chromosomal spatial correlations across chromosome pairs and genome-wide interactions.
• Functional Group Analysis: Detects functional groups of genes that exhibit strong spatial dependencies in expression.

Scientific Applications:

• 3D genome–expression coupling: Quantifies how 3D chromosome structures correlate with gene expression variation using integrated HiC and RNA-seq data.
• Chromatin organization studies: Characterizes intra-chromosomal dependencies within chromatin loops and TADs and evaluates TAD boundary insulation on expression.
• Inter-chromosomal interaction analysis: Assesses genome-wide and chromosome-pair spatial correlations to reveal inter-chromosomal regulatory relationships.
• Functional regulation discovery: Identifies groups of protein-coding genes with coordinated expression driven by spatial proximity.

Methodology:

R package implementation of a Poisson Hierarchical Markov Random Field (PHMRF) for count data that integrates RNA-seq and HiC inputs, applies Bayesian inference to estimate spatial dependencies among protein-coding genes, and computes the Spatial Interaction Estimate (SIE).