BayesMD

BayesMD applies a fully Bayesian framework to discover sequence motifs and infer transcription factor binding sites by integrating multiple layers of biological priors and advanced statistical inference.

Key Features:

• Modular biological priors: Uses a mixture of Dirichlet priors over nucleotide probabilities trained on transcription factor (TF) databases, organism-specific background priors, and positional priors that integrate conservation, local sequence complexity, nucleosome occupancy, and assumptions about the number of occurrences.
• Bayesian inference: Combines exact marginalization of multinomial parameters with sampling of binding-site positions.
• Parallel tempering: Employs parallel tempering to improve sampling convergence and robustness.
• Post-analysis motif identification: Identifies candidate motifs by searching among motifs containing frequently occurring sites and assesses motif significance using marginal probabilities rather than maximum a posteriori inference.
• Benchmarking and validation: Validated and benchmarked against other methods using both real and artificial datasets.

Scientific Applications:

• Microarray analysis: Applied to motif discovery in microarray experiments.
• ChIP-chip experiments: Used to detect transcription factor binding sites from ChIP-chip data.
• Ditag sequencing: Applied to motif discovery in ditag sequencing datasets.
• CAGE data analysis: Applied to motif discovery in CAGE datasets.
• Transcription factor binding site and regulatory element analysis: Used to uncover motifs underlying gene regulation and expression.

Methodology:

Training a Dirichlet mixture prior on TF databases; constructing organism-specific background sequence models; incorporating positional priors that integrate conservation, local sequence complexity, nucleosome occupancy, and occurrence-number assumptions; performing exact marginalization of multinomial parameters and sampling binding-site positions with parallel tempering; identifying candidate motifs by searching frequently occurring sites and assessing significance via marginal probabilities.