Stochastic EM-based TFBS motif discovery with MITSU

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The Expectation-Maximization (EM) algorithm is foundational for transcription factor binding site (TFBS) motif discovery.
• Stochastic EM (sEM) offers potential improvements over standard EM but is underexplored in motif discovery.

Purpose of the Study:

• To introduce MITSU, a novel motif discovery algorithm leveraging sEM.
• To address limitations of existing EM-based motif discovery methods.

Main Methods:

• Developed MITSU (Motif discovery by ITerative Sampling and Updating) algorithm.
• Integrated sEM with an advanced, unconstrained likelihood function approximation.
• Evaluated performance on synthetic and prokaryotic TFBS datasets.

Main Results:

• MITSU demonstrated superior performance compared to standard EM and an alternative sEM algorithm.
• The algorithm showed particular strength in site-level positive predictive value.
• Quantitative evaluation confirmed effectiveness on realistic datasets.

Conclusions:

• MITSU represents a significant advancement in TFBS motif discovery.
• The novel combination of sEM and improved likelihood approximation offers enhanced predictive accuracy.