VOMBAT

VOMBAT predicts transcription factor binding sites and DNA sequence motifs using variable order Markov models and variable order Bayesian trees to capture position-specific and context-dependent nucleotide dependencies.

Key Features:

• Variable Order Models: Implements variable order Markov models and Variable Order Bayesian Network (VOBN) models that adapt position-specific nucleotide dependency subsets to capture statistical dependencies, including non-adjacent positions.
• Enhanced Accuracy: Demonstrates improved TFBS discrimination over position weight matrices (PWMs), fixed-order Markov models, and conventional Bayesian networks, exemplified by sigma-70 binding sites in Escherichia coli with a mean true-positive rate of 47.56% versus 44.39% for conventional models.
• Training, Prediction, and Evaluation: Trains variable order Markov models and Bayesian trees on datasets with annotated binding sites and genomic background sequences, predicts putative DNA binding sites in genomic sequences, and performs cross-validation experiments to evaluate model combinations and parameter settings.
• Scalability and Performance: Supports genome-wide DNA binding site predictions and extensive cross-validation experiments, operating on a Linux cluster with 150 processors for high-throughput computation.

Scientific Applications:

• Gene regulation and transcription factor dynamics: Identifies transcription factor binding sites to support studies of gene regulation and transcription factor dynamics.
• Regulatory mechanism elucidation: Facilitates understanding of regulatory mechanisms governing gene expression through accurate TFBS identification.
• Genome-wide analyses: Supports genome-wide prediction of DNA binding sites and large-scale analyses relevant to genome-wide association studies.
• Comparative genomics: Enables comparative genomics research by providing motif predictions across genomes.

Methodology:

Uses variable order Markov models and variable order Bayesian trees (Variable Order Bayesian Networks, VOBNs); trains on datasets with annotated binding sites and genomic background sequences; performs cross-validation experiments and direct comparisons to PWMs, fixed-order Markov models, and conventional Bayesian networks, with evaluation reported on sigma-70 binding sites in Escherichia coli.