Predicting transcription factor site occupancy using DNA sequence intrinsic and cell-type specific chromatin features

Area of Science:

• Genomics
• Molecular Biology
• Computational Biology

Background:

• Transcription factor (TF) recruitment to DNA target sites is vital for gene regulation.
• DNA sequence binding affinity alone poorly predicts in vivo TF occupancy and cell-type specificity.
• Chromatin accessibility, nucleosome occupancy, and histone modifications significantly influence TF binding in vivo.

Purpose of the Study:

• To develop and apply machine-learning models to predict TF target site occupancy.
• To assess the relative importance of DNA sequence-intrinsic and chromatin features in TF recruitment.
• To understand the mechanisms of cell-type specific TF binding.

Main Methods:

• Utilized ENCODE consortium data for five transcription factors (CTCF, JunD, REST, GABP, USF2) across multiple cell types.
• Analyzed predicted TF binding sites and cell-type specific peak lists.
• Computed sequence-intrinsic features and experimental chromatin features (histone modifications, DNase I hypersensitivity).
• Employed machine learning for binary classification and regression to predict TF site occupancy and binding strength.

Main Results:

• Feature importance varied significantly across the five TFs.
• Position weight matrix (PWM) scores were important for CTCF and REST, but not JunD or USF2.
• Chromatin accessibility and active histone marks effectively predicted binding for most TFs, except REST.
• Repressive histone marks and structural DNA parameters showed limited predictive value.

Conclusions:

• Developed a computational framework to analyze DNA-intrinsic and chromatin features for cell-type specific TF binding.
• The framework provides insights into transcription regulatory processes.
• This methodology can be applied to larger datasets for further understanding of gene regulation.