Modelling time course gene expression data with finite mixtures of linear additive models

Area of Science:

• Statistics
• Computational Biology
• Machine Learning

Background:

• Finite mixture models are widely used for clustering and density estimation.
• Linear additive models offer flexibility in capturing complex relationships.
• Regularization techniques are crucial for stabilizing parameter estimation in high-dimensional or complex models.

Purpose of the Study:

• To present a novel model class: finite mixtures of linear additive models.
• To introduce regularized likelihood methods for estimating component-specific parameters.
• To demonstrate the advantages of regularization in spline-based regression within mixture models.

Main Methods:

• Development of a finite mixture model framework for linear additive components.
• Application of regularized likelihood estimation for parameter inference.
• Automatic selection of degrees of freedom for splines within each component.
• Evaluation using simulation studies and a real-world yeast cell cycle gene expression dataset.

Main Results:

• Regularized estimation reduces sensitivity to the pre-specified maximum degrees of freedom for splines.
• Automatic degree of freedom selection enhances model adaptability for each component.
• The model demonstrates effective performance on both artificial and biological data.

Conclusions:

• The proposed regularized finite mixture of linear additive models offers a robust and flexible approach to data analysis.
• This method provides automatic control over model complexity, simplifying practical application.
• The findings are relevant for statistical modeling, machine learning, and analysis of biological time-series data.