Finding sequence motifs with Bayesian models incorporating positional information: an application to transcription...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Biologically active sequence motifs, such as transcription factor binding sites (TFBSs), often exhibit positional preferences relative to genomic landmarks like transcription start sites (TSSs).
• Existing motif identification programs often implicitly or inadequately model this positional information, limiting their general applicability.

Purpose of the Study:

• To develop and evaluate A-GLAM, a computer program that integrates sequence and positional information using a Bayesian model for enhanced motif discovery.
• To assess the impact of positional information on the accuracy and robustness of motif prediction.

Main Methods:

• A-GLAM was developed incorporating a Bayesian model to systematically combine sequence and positional data.
• A-GLAM's predictions were compared with and without positional information on human TFBS datasets spanning [-2000, 0] bases upstream of known TSSs.
• Performance was evaluated using rigorous statistical analysis and cross-validation, including analyses with progressively truncated sequence intervals.

Main Results:

• Positional information significantly improved the prediction accuracy of sequence motifs.
• A-GLAM demonstrated robustness against minor parameter misspecifications in cross-validation.
• While the benefit decreased with shorter sequence intervals, positional information never significantly harmed motif prediction.

Conclusions:

• A probabilistic model, like the one in A-GLAM, offers a superior and more robust strategy for identifying biologically active motifs with positional preferences compared to sequence truncation.
• This probabilistic approach is particularly advantageous when the precise positional preferences of motifs are not well-defined.