MIT researchers have developed ChromoGen, a GenAI model that rapidly predicts 3D chromatin structures, improving genome analysis efficiency.

MIT ChromoGen overcomes the limitations of traditional chromatin mapping techniques, which are slow and resource-intensive. Standard methods like Hi-C require weeks to analyze a single cell, limiting researchers’ ability to study genome organization. ChromoGen, leveraging deep learning and GenAI, generates thousands of genome structures within minutes, making large-scale studies feasible. By integrating deep learning with GenAI, ChromoGen enables large-scale studies of gene regulation and disease mechanisms that were previously impractical due to time and cost constraints.

Trained on over 11 million chromatin conformations, generates thousands of structures in minutes, dramatically accelerating research. The model operates in two parts: a deep learning component trained to interpret DNA sequences and chromatin accessibility data, and a GenAI engine that predicts chromatin conformations based on physical constraints. This dual-system approach enables high accuracy in mapping chromatin folding across different cell types. Unlike traditional methods, ChromoGen can generalize to unseen cell types using only DNA sequences and widely available accessibility data.

By enabling large-scale, cost-effective genome analysis, ChromoGen opens new possibilities for studying gene regulation and disease mechanisms. The model’s ability to generate multiple conformations for a given sequence provides a comprehensive view of genome folding, helping to uncover how structural variations influence cellular functions. Researchers can also analyze how DNA mutations alter chromatin structures, shedding light on their role in genetic diseases.

Researchers can now investigate chromatin variations between cell types, track genome structural changes over time, and explore how DNA mutations impact gene expression. With its ability to generate high-dimensional chromatin structures rapidly, ChromoGen marks a major step in AI-driven genomic research.