BayesMD: flexible biological modeling for motif discovery

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying transcription factor (TF) binding sites is crucial for understanding gene regulation.
• Existing motif discovery methods may not fully leverage diverse biological prior knowledge.

Purpose of the Study:

• To introduce BayesMD, a novel Bayesian Motif Discovery model.
• To enhance motif discovery by incorporating multiple types of biological a priori knowledge.

Main Methods:

• Utilizing a mixture of Dirichlets as a prior for nucleotide probabilities in binding sites, trained on TF databases.
• Developing organism-specific priors for background sequences and a prior for binding site positions incorporating conservation and nucleosome occupancy.
• Employing Bayesian inference with exact marginalization and parallel tempering sampling for robust results.

Main Results:

• BayesMD integrates diverse biological priors (TF binding site properties, organism-specific background, positional information) into a modular framework.
• The model uses Bayesian inference with advanced sampling techniques for accurate motif discovery.
• Candidate motifs are identified based on high marginal probability, avoiding maximum a posteriori inference for direct significance assessment.

Conclusions:

• BayesMD offers a robust and flexible framework for motif discovery by integrating various biological priors.
• The model demonstrates strong performance through benchmarking against other methods on real and artificial datasets.
• Associated resources including a prediction server, software, and data are publicly available.