Modeling within-motif dependence for transcription factor binding site predictions

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• The position-specific weight matrix (PWM) model is standard for describing transcription factor binding site motifs.
• Recent findings indicate interdependence among positions within DNA binding sites, challenging the independent position assumption of PWMs.

Purpose of the Study:

• To extend the PWM model by incorporating correlated positions to improve motif discovery.
• To develop a computational algorithm that leverages these correlations for more accurate identification of transcription factor binding sites.

Main Methods:

• Extended the position-specific weight matrix (PWM) model to include correlated positions.
• Designed a Markov chain Monte Carlo algorithm for sampling within the extended model space.
• Integrated model sampling with the Gibbs sampling framework for de novo motif discovery.

Main Results:

• Approximately 25% of transcription factor binding motifs exhibit significant within-site position correlations.
• Considering correlated positions improved motif models for 80% of those identified with correlations.
• The new algorithm accurately infers correlated position pairs and enhances precision in finding transcription factor binding sites compared to standard Gibbs sampling.

Conclusions:

• Accounting for positional interdependence in transcription factor binding sites significantly improves motif discovery.
• The developed algorithm offers a more precise and accurate method for identifying transcription factor binding sites, especially those with correlated positions.