PyMEGABASE

PyMEGABASE predicts chromosomal compartments and subcompartments from local epigenomic profiles to provide high-resolution structural annotation of genomic loci without requiring DNA-proximity-ligation (Hi-C) data.

Key Features:

• Epigenome-Based Prediction: Classifies genomic loci into compartments and subcompartments using histone modifications and other epigenetic signals instead of Hi-C or other DNA-proximity-ligation experiments.
• Maximum-Entropy Neural Network: Implements a maximum-entropy-based neural network model to infer structural classes from epigenomic markers.
• High-Resolution Annotation: Generates compartment and subcompartment predictions at resolutions up to 5 kilobases (kbp).
• Cross-Species Generalization: Trained on human cell data and validated to predict chromosomal compartments in other species, including mouse.
• Model Interpretability: Enables analysis of trained parameters to quantify contributions of specific histone modifications and epigenetic features to subcompartment classification.
• OpenMiChroM Integration: Produces outputs compatible with OpenMiChroM for three-dimensional genome structure modeling.

Scientific Applications:

• Genome Architecture and Gene Regulation: Annotates (sub)compartments across diverse human cell types, including ENCODE datasets, to investigate relationships among chromosomal organization, epigenomic state, gene expression, and cellular differentiation.

Methodology:

PyMEGABASE applies a maximum-entropy-based neural network trained on human epigenomic datasets to predict chromosomal compartment and subcompartment annotations from local histone modification and other epigenetic marker profiles.