The challenge of chromatin model comparison and validation: A project from the first international 4D Nucleome Hackathon.

Abstract

The computational modeling of chromatin structure is highly complex due to the hierarchical organization of chromatin, which reflects its diverse biophysical principles, as well as inherent dynamism, which underlies its complexity. Chromatin structure modeling can be based on diverse approaches and assumptions, making it essential to determine how different methods influence the modeling outcomes. We conducted a project at the NIH-funded 4D Nucleome Hackathon on March 18-21, 2024, at The University of Washington in Seattle, USA. The hackathon provided an amazing opportunity to gather an international, multi-institutional and unbiased group of experts to discuss, understand and undertake the challenges of chromatin model comparison and validation. Here we give an overview of the current state of the 3D chromatin field and discuss our efforts to run and validate the models. We used distance matrices to represent chromatin models and we calculated Spearman correlation coefficients to estimate differences between models, as well as between models and experimental data. In addition, we discuss challenges in chromatin structure modeling that include: 1) different aspects of chromatin biophysics and scales complicate model comparisons, 2) large diversity of experimental data (e.g., population-based, single-cell, protein-specific) that differ in mathematical properties, heatmap smoothness, noise and resolutions complicates model validation, 3) expertise in biology, bioinformatics, and physics is necessary to conduct comprehensive research on chromatin structure, 4) bioinformatic software, which is often developed in academic settings, is characterized by insufficient support and documentation. We also emphasize the importance of establishing guidelines for software development and standardization.