EPIVAN

EPIVAN predicts long-range enhancer–promoter interactions (EPIs) from genomic sequence data using vector representations and attention-based deep neural networks.

Key Features:

• Sequence-Based EPI Prediction: Predicts enhancer–promoter interactions using only genomic sequence information without requiring additional genomic or epigenomic datasets.
• Pre-trained DNA Vector Encoding: Encodes enhancer and promoter sequences using pre-trained DNA vectors that capture sequence-level characteristics.
• Hybrid Deep Learning Architecture: Utilizes one-dimensional convolutional neural networks (CNNs) and gated recurrent units (GRUs) to extract local and long-range sequence features.
• Attention Mechanism: Applies an attention layer to emphasize informative sequence features that contribute to enhancer–promoter interaction prediction.

Scientific Applications:

• Enhancer–Promoter Interaction Analysis: Identifies long-range regulatory interactions between enhancers and promoters in genomic sequences.
• Regulatory Genomics Studies: Supports investigation of gene regulation mechanisms involved in development and cellular function.
• Cross-Cell-Line EPI Prediction: Enables prediction of enhancer–promoter interactions in cell lines lacking extensive genomic or epigenomic data.

Methodology:

EPIVAN encodes enhancer and promoter sequences using pre-trained DNA vectors, extracts sequence features with one-dimensional CNNs and GRUs, and applies an attention mechanism to prioritize informative features for predicting enhancer–promoter interactions.