Logos: a modular bayesian model for de novo motif detection

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Higher eukaryotic organisms exhibit complex motif organization, challenging traditional motif detection methods.
• Accurate de novo motif detection requires modeling intricate dependencies within and between motifs, incorporating biological prior knowledge.

Purpose of the Study:

• To present LOGOS, an integrated Local and Global motif Sequence model for biopolymer sequences.
• To provide a flexible framework for developing, modularizing, extending, and computing motif models for complex sequence analysis.

Main Methods:

• LOGOS comprises two submodels: HMDM (local alignment model) and HMM (global motif distribution model).
• Incorporates biological prior knowledge and positional dependencies within local motif structures.
• Employs an empirical Bayesian framework for parameter fitting and a variational EM algorithm for de novo motif detection.

Main Results:

• LOGOS outperforms existing models by accounting for biological priors and dependencies in motif structures and occurrences.
• Demonstrates superior sensitivity, specificity, flexibility, and extensibility on test data and genomic sequences.
• Effective for cis-regulatory sequence analysis in yeast and Drosophila.

Conclusions:

• LOGOS offers a principled and extensible framework for advanced motif modeling in biopolymer sequence analysis.
• The integrated local and global approach significantly improves de novo motif detection accuracy.
• LOGOS represents a substantial advancement over models that neglect crucial biological information.