Penalized likelihood for sparse contingency tables with an application to full-length cDNA libraries.

Area of Science:

• Computational biology
• Systems biology
• Statistical genetics

Background:

• Joint analysis of categorical variables is crucial for systems biology.
• Log-linear models traditionally capture complex interactions but struggle with high-dimensional, sparse biological data.
• Analyzing full-length cDNA libraries of alternatively spliced genes requires robust methods for interaction detection.

Purpose of the Study:

• To develop and evaluate methods for model selection and parameter estimation in log-linear models for sparse contingency tables.
• To address challenges in analyzing high-dimensional biological data with potential zeros in contingency tables.
• To identify complex interaction patterns among categorical variables in biological networks.

Main Methods:

• Developed an efficient l1-penalization approach, extending the Lasso algorithm for log-linear models.
• Proposed regularization methods for parameter estimation and model selection in sparse contingency tables.
• Compared the proposed methods against other procedures using simulation studies.

Main Results:

• The proposed l1-penalization method efficiently handles sparse contingency tables common in computational biology.
• Demonstrated the effectiveness of the regularization methods on contingency tables from full-length cDNA libraries.
• Showcased the ability to perform model selection and parameter estimation in challenging biological datasets.

Conclusions:

• Regularization methods successfully detect complex interaction patterns among categorical variables.
• The developed algorithms are applicable to a broad range of biological problems involving categorical data analysis.
• Provides a robust framework for analyzing interactions in systems biology and genetic studies.