DeepMotifSyn

DeepMotifSyn synthesizes heterodimeric transcription factor (TF) DNA-binding motifs from pairs of monomeric motifs to predict TF cooperativity in gene regulation.

Key Features:

• Heterodimeric Motif Generator: A U-Net-based neural network generates potential heterodimeric motifs from aligned monomeric motif pairs.
• Evaluator: A machine-learning model assesses generated heterodimeric motifs based on sequence features to assign confidence scores.
• Flexibility in synthesis: Generates multiple heterodimeric motifs with varying orientations and spacing configurations.
• Performance and benchmarking: Systematic evaluations using CAP-SELEX data demonstrate accuracy that surpasses existing state-of-the-art predictors.
• Scalability for large search spaces: Synthesizes and evaluates many candidate heterodimeric motifs to address the large number of potential TF interactions.

Scientific Applications:

• Identification of cooperative TF interactions: Predicts heterodimeric motifs to identify potential cooperative transcription factor pairs.
• Prioritization for experimental validation: Ranks candidate heterodimeric TF pairs to guide targeted experimental testing.
• Interpretation of composite binding data: Aids interpretation and modeling of CAP-SELEX composite DNA sites and regulatory networks involved in gene expression and cellular processes.

Methodology:

DeepMotifSyn comprises a U-Net-based heterodimeric motif generator that produces candidate motifs from aligned monomeric motif pairs and a machine-learning evaluator that scores those candidates based on sequence features, with performance benchmarked using CAP-SELEX data.