Deep neural networks identify sequence context features predictive of transcription factor binding.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Transcription factors (TFs) regulate gene expression by binding to specific DNA sequences.
• While TF motifs are abundant, only a fraction are actively bound, necessitating methods to predict functional binding sites.

Purpose of the Study:

• To develop and apply a machine learning framework for predicting TF binding at specific DNA motif instances.
• To identify and interpret sequence context features that are most predictive of TF binding.

Main Methods:

• Utilized convolutional neural network architectures and model interpretation techniques.
• Developed a framework to predict TF binding for 38 TFs in a lymphoblastoid cell line.
• Scored context sequence importance at base-pair resolution.

Main Results:

• The choice of training data significantly impacts classification accuracy and feature importance.
• Identified key context features, including open chromatin, predictive of TF binding.
• The framework successfully predicts TF binding based on sequence context.

Conclusions:

• The developed framework provides novel insights into features governing TF binding.
• This approach can inform future deep learning applications for interpreting non-coding genetic variants and understanding gene regulation.