Model selection and sensitivity analysis for sequence pattern models.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Modeling

Background:

• Motif discovery is crucial for understanding gene regulation.
• Accurate model selection is a key challenge in motif discovery.
• Existing methods may lack robustness to prior assumptions.

Purpose of the Study:

• To propose a novel maximal a posteriori (MAP) criterion for model selection in motif discovery.
• To evaluate the asymptotic correctness of the MAP criterion in predicting model size.
• To assess the robustness of the MAP criterion to variations in prior specifications.

Main Methods:

• Development of a maximal a posteriori (MAP) criterion.
• Asymptotic analysis to determine conditions for correct model size prediction.
• Sensitivity analysis to evaluate robustness against prior hyper-parameter choices.

Main Results:

• The proposed MAP criterion provides asymptotically correct model size predictions under specific conditions.
• The MAP criterion demonstrates robustness to different prior specifications.
• Guidelines for selecting prior hyper-parameters are established based on sensitivity analysis.

Conclusions:

• The MAP criterion is a reliable tool for model selection in motif discovery.
• Understanding prior sensitivity enhances the applicability of the MAP criterion.
• This work provides a principled approach to motif model selection and hyper-parameter tuning.