GraphReg

GraphReg models gene regulation by integrating 3D chromatin interactions from chromosome conformation capture assays with 1D epigenomic data or genomic DNA sequence to link distal enhancers to target genes and predict effects of noncoding variation on gene expression.

Key Features:

• Graph Attention Networks (GATs): GraphReg employs graph attention networks to capture the connectivity of distal regulatory elements up to 2 megabases across the genome for gene regulatory modeling.
• Integration of Chromosome Conformation Capture Data: It integrates chromosome conformation capture data to incorporate 3D chromatin interactions when predicting gene expression levels with greater precision than existing deep learning models.
• Enhancer Identification and Functional Validation: Feature attribution identifies functional enhancers associated with genes, with validations via CRISPR interference (CRISPRi) combined with FlowFISH and TAP-seq, and reported improved performance over convolutional neural networks (CNNs) and the activity-by-contact model.
• Transcription Factor Target Prediction: Sequence-based analyses predict direct transcription factor (TF) targets, validated through in silico ablation of TF binding motifs and CRISPRi TF knockout experiments.

Scientific Applications:

• Gene expression prediction: Predicting gene expression levels from combined 3D chromatin interaction and 1D epigenomic or sequence data.
• Enhancer–gene linking: Linking distal enhancers to target genes to map regulatory interactions across up to 2 Mb.
• Interpretation of noncoding variation: Interpreting the regulatory impact of noncoding genetic variation on target gene expression.
• TF target identification: Identifying direct transcription factor targets for experimental follow-up using sequence-based motif analyses and in silico ablation.

Methodology:

GraphReg uses deep learning with graph attention networks to process 3D chromatin interaction data alongside 1D epigenomic signals or genomic DNA sequence, employs feature attribution for enhancer identification, and uses in silico TF motif ablation for TF target prediction.