PEPR-GNN: Perturbation-Enhancer-Promoter-RNA Graph Neural Networks for Multiome Perturb-Seq modeling of regulomes

Area of Science:

• * Molecular Biology
• * Computational Biology
• * Genomics

Background:

• * Cellular reprogramming involves complex gene expression changes driven by regulatory elements.
• * Existing computational methods lack explicit modeling of regulatory interactions during reprogramming.
• * Understanding these interactions is crucial for controlling cellular fate decisions.

Purpose of the Study:

• * To develop a computational framework, PEPR-GNN (Perturbation-Enhancer-Promoter-RNA Graph Neural Network), for modeling regulome responses to complex genetic perturbations.
• * To analyze gene regulatory relationships and identify distinct regulomes based on reprogramming responses.
• * To enable in silico prediction of how enhancer modifications can modulate gene responses.

Main Methods:

• * Combinatorial reprogramming using cardiac transcription factors.
• * Multiome Perturb-Seq to simultaneously measure perturbations, open chromatin, and gene expression at the single-cell level.
• * Development and application of the PEPR-GNN framework, a graph neural network model.

Main Results:

• * PEPR-GNN successfully models regulome responses to complex perturbations by statistically associating gene regulatory relationships.
• * Identified distinct regulomes: easily reprogrammed cardiac genes, difficult-to-reprogram fibroblast genes, and context-specific genes.
• * Demonstrated the utility of PEPR-GNN for in silico modeling to tune gene responses via enhancer modifications.

Conclusions:

• * PEPR-GNN provides a novel enhancer-aware regulome model to effectively capture complex cellular responses to perturbations.
• * The framework leverages causal perturbation data to enhance the understanding of gene regulation.
• * PEPR-GNN facilitates predictive modeling for precise control over cellular reprogramming outcomes.