Chromatin structures from integrated AI and polymer physics model.

Area of Science:

• Genomics
• Computational Biology
• Biophysics

Background:

• The three-dimensional (3D) organization of the genome is fundamental to regulating gene expression and cellular processes.
• Accurate characterization of genome structure is essential for understanding these biological mechanisms.
• Experimental determination of 3D genome structure is complex, necessitating robust computational modeling.

Purpose of the Study:

• To develop an efficient and accurate computational approach for estimating 3D chromatin structure from indirect experimental measures.
• To integrate a particle-based polymer model with molecular simulation and machine learning (ML).
• To leverage graph neural networks (GNNs) for extracting polymer model parameters from Hi-C data.

Main Methods:

• A particle-based chromatin polymer model was employed.
• Molecular simulation techniques were utilized.
• A graph neural network (GNN) was developed to extract interaction parameters from Hi-C data.
• The GNN was trained primarily on simulated data derived from the polymer model.

Main Results:

• The developed approach accurately estimates chromatin structures across all chromosomes.
• The method demonstrates efficacy across multiple experimental cell lines.
• The GNN model achieved high accuracy despite being trained predominantly on simulated data.

Conclusions:

• The combined physical modeling and ML framework provides a powerful tool for 3D genome structure prediction.
• This approach enables accurate and high-throughput chromatin structure estimation from Hi-C data.
• The methodology offers a generalizable framework for integrating diverse biological data modalities for structural modeling.