Novel Machine Learning Model For Transcription Factor (TF) Binding Prediction

Abstract

Transcription factors (TFs) are fundamental to gene regulation. Many TFs selectively bind to DNA at distinctive sequence motifs, providing the sequence specificity necessary to control gene regulation mechanisms. The study of TF binding in a cellular environment is therefore essential to our understanding of growth, development, differentiation, evolution, and disease. Most current computational techniques that model and predict the DNA binding of TF rely upon only sequence-specific binding data, ignoring the cellular context. In our project, we aimed to improve the prediction of TF binding by incorporating data from different sources, such as ChIP-seq, ATAC-seq, DNase-seq, and histone marks, into a combined dataset. We trained both traditional and complex models on this combined dataset to study TF binding. We observed that the addition of epigenomic data significantly improved the performance of all the models. This suggests that incorporating multiple types of epigenomic data can enhance our ability to predict TF binding and improve our understanding of gene regulation mechanisms in vivo.