Occupancy classification of position weight matrix-inferred transcription factor binding sites

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Predicting transcription factor binding sites (TFBS) solely from sequence data is challenging and prone to errors.
• Machine learning (ML) models integrating environmental data, like chromatin feature proximity, enhance TFBS prediction accuracy.
• Evaluating Bayesian Network (BN) and Support Vector Machine (SVM) ML techniques for TFBS occupancy classification.

Purpose of the Study:

• To assess the efficacy of ML techniques, specifically BN and SVM, for predicting transcription factor binding site (TFBS) occupancy.
• To identify key environmental features, such as chromatin modifications, that improve TFBS prediction.
• To compare the performance of BN and SVM classifiers on TFBS data.

Main Methods:

• Utilized Bayesian Network (BN) and Support Vector Machine (SVM) algorithms.
• Trained and tested classifiers on four distinct Transcription Factor Binding Site (TFBS) datasets.
• Analyzed feature importance, focusing on distances to chromatin features and between modifications.

Main Results:

• Both BN and SVM classifiers demonstrated strong performance on TFBS prediction, including cross-classification experiments.
• Distances to histone modification islands and between modifications proved effective in predicting TFBS occupancy.
• Bayesian network classifiers consistently outperformed SVM classifiers in the conducted experiments.

Conclusions:

• Machine learning techniques, particularly those using chromatin feature distances, are effective for TFBS occupancy classification.
• Cross-classification of TFBS is feasible, indicating potential for developing generalizable occupancy classifiers applicable to multiple transcription factors.