DeepSELEX

DeepSELEX infers transcription factor (TF) DNA-binding preferences from high-throughput Systematic Evolution of Ligands by Exponential Enrichment (HT-SELEX) data using multi-class Convolutional Neural Networks (CNNs).

Key Features:

• Deep Learning Integration: Employs multi-class Convolutional Neural Networks (CNNs) to capture intricate sequence patterns indicative of TF binding.
• Sequential Data Analysis: Learns from changes in DNA sequence abundance across multiple HT-SELEX enrichment cycles to model selection dynamics.
• Performance Superiority: Outperforms existing computational methods for in vitro binding prediction and matches state-of-the-art performance for in vivo predictions.
• Biologically Relevant Insights: Analysis of model parameters reveals learned features that correspond to biologically relevant determinants of TF–DNA interactions.

Scientific Applications:

• Gene Regulation Studies: Investigating the roles of specific TFs in regulating gene expression by predicting their binding preferences.
• Disease Research: Identifying aberrant TF binding sites associated with diseases such as cancer.
• Drug Development: Informing strategies to target TF–DNA interactions for therapeutic development.

Methodology:

DeepSELEX trains multi-class CNNs on sequence information obtained from multiple HT-SELEX enrichment cycles and models spatial hierarchies in DNA sequences by learning from the sequential changes observed during SELEX.