A model-based clustering algorithm with covariates adjustment and its application to lung cancer stratification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clustering is crucial for data analysis, revealing hidden patterns and generating hypotheses.
• Covariates can obscure true clustering structures in empirical data, leading to misinterpretations.
• Standard clustering may group subjects by confounding factors (e.g., age) rather than disease status.

Purpose of the Study:

• To introduce CEM-Co, a model-based clustering algorithm designed to mitigate the influence of undesirable covariates.
• To demonstrate the efficacy of CEM-Co in identifying meaningful subgroups within complex biological datasets.

Main Methods:

• Developed CEM-Co, a novel model-based clustering algorithm.
• Applied CEM-Co to a gene expression dataset from 129 stage I non-small cell lung cancer patients.
• Compared CEM-Co's performance against standard clustering algorithms.

Main Results:

• CEM-Co successfully identified a subgroup of lung cancer patients with a poorer prognosis.
• Standard clustering algorithms failed to detect this clinically relevant subgroup.
• CEM-Co effectively removed or minimized the confounding effects of covariates in the clustering process.

Conclusions:

• CEM-Co offers a robust approach for covariate-adjusted clustering in biological data analysis.
• This method enhances the ability to discover clinically significant subgroups, improving prognostic identification.
• CEM-Co has potential applications in various fields dealing with covariate-influenced data.