ChromoGen: Diffusion model predicts single-cell chromatin conformations.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• High-throughput sequencing and imaging reveal cell-specific chromatin structure variations.
• Characterizing this heterogeneity is challenging due to labor-intensive experiments.

Purpose of the Study:

• Introduce ChromoGen, an AI-driven generative model.
• Enable efficient prediction of 3D single-cell chromatin conformations de novo.
• Facilitate systematic investigation of chromatin organization and heterogeneity.

Main Methods:

• Developed ChromoGen, a generative AI model.
• Trained the model on existing chromatin conformation data.
• Validated predictions against experimental results at single-cell and population levels.
• Tested model transferability to new cell types using DNA sequence and DNase-seq data.

Main Results:

• ChromoGen accurately predicts single-cell chromatin conformations with region and cell type specificity.
• Generated conformations align with experimental data.
• The model demonstrates successful transfer learning to novel cell types.
• Achieved high accuracy at a low computational cost.

Conclusions:

• ChromoGen offers an economical and efficient approach to study single-cell chromatin organization.
• The model overcomes limitations of experimental characterization of chromatin heterogeneity.
• Enables broader access to chromatin structure data across diverse cell types.