A model selection approach to discover age-dependent gene expression patterns using quantile regression models.

Area of Science:

• Genomics
• Computational Biology
• Aging Research

Background:

• Understanding mammalian aging mechanisms at the molecular level remains a challenge.
• Existing methods for identifying age-dependent gene expression patterns (differential expression and differential variability) often use linear regression, which is sensitive to outliers and non-linear patterns.
• A more robust and flexible approach is needed to accurately capture diverse age-related gene expression changes.

Purpose of the Study:

• To develop and present a novel model selection approach for discovering genes with linear or non-linear age-dependent expression patterns from microarray data.
• To identify differentially expressed (DE) and differentially variable (DV) genes using quantile regression.
• To demonstrate the robustness and applicability of this new approach in aging research.

Main Methods:

• Utilized quantile regression models (constant, linear, piecewise linear) to fit gene expression profiles and select the best-fitting model for identifying DE genes.
• Employed quantile regression models (non-DV and DV) to identify DV genes.
• Applied a model selection strategy to compare different models and identify the most appropriate one for each gene's expression pattern.

Main Results:

• The proposed quantile regression-based model selection approach is significantly more robust than standard linear regression for detecting age-dependent patterns.
• Analysis of human brain aging datasets revealed numerous biologically relevant gene expression patterns, including previously overlooked DV patterns.
• The method successfully identified both linear and non-linear age-dependent expression patterns.

Conclusions:

• A novel application of quantile regression models and a model selection approach effectively identifies genes with linear or non-linear age-dependent expression patterns.
• This method offers a robust alternative to traditional null hypothesis testing approaches for DE and DV gene identification.
• The approach is broadly applicable to various aging and time-series microarray data analysis tasks.