CBCR

CBCR reconstructs three-dimensional chromosome structures from Hi-C contact data to model chromosomal spatial configuration and output xyz coordinates in .pdb format.

Key Features:

• Curriculum Learning Strategy: Categorizes Hi-C contacts by contact frequency and progressively trains the model using a curriculum learning approach.
• Distance-Restraint Algorithm: Implements a distance-restraint-based algorithm using a modified Gaussian maximum likelihood function that scales probabilities according to feature importance.
• Performance and Validation: Evaluated on simulated and actual Hi-C datasets and benchmarked against FISH, HiChIP, and ChIA-PET.
• Computational Efficiency and Robustness: Accelerates optimization convergence, reduces computational cost, and maintains robust performance across varying data resolutions with improved accuracy for high-resolution datasets.

Scientific Applications:

• Chromosomal organization analysis: Provides 3D models to study chromosomal organization and spatial configuration from Hi-C interaction frequencies.
• Functional genomics: Supports investigations of gene regulation, DNA replication, and cellular function by supplying reconstructed chromosome structures.
• Spatial genomics and disease research: Enables exploration of spatial dynamics of genomic elements and investigation of complex biological processes and disease mechanisms.

Methodology:

Categorizes Hi-C contact frequencies into classes; progressively trains a distance-restraint-based algorithm via curriculum learning; employs a modified Gaussian maximum likelihood function that scales probabilities by feature importance; outputs xyz coordinates and .pdb files; evaluated using simulated and actual Hi-C datasets and compared to FISH, HiChIP, and ChIA-PET.