Supervised clustering of high-dimensional data using regularized mixture modeling.

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• Disease etiology is often heterogeneous, complicating the identification of relationships between genetic variations and clinical presentations.
• Understanding high-dimensional genetic data and its link to phenotypes requires methods that account for subject heterogeneity.

Purpose of the Study:

• To develop a novel supervised clustering algorithm to address challenges in studying heterogeneous relationships between high-dimensional genetic features and phenotypes.
• To improve computational efficiency and biological interpretability in genetic association studies.

Main Methods:

• Proposed a novel supervised clustering algorithm, component-wise sparse mixture regression (CSMR), utilizing a penalized mixture regression model.
• Adapted the classification expectation maximization algorithm for a supervised clustering approach.
• Evaluated CSMR on simulated benchmark datasets and a drug sensitivity dataset.

Main Results:

• CSMR accurately identifies feature subspaces that explain response variables on simulated data.
• The algorithm demonstrated superior performance compared to baseline methods in both simulated and real-world datasets.
• CSMR effectively recapitulated distinct subgroups in cell lines based on drug response, revealing coping mechanisms.

Conclusions:

• CSMR is a powerful big data analysis tool for resolving the complexity of translating clinical disease representations to underlying causes.
• The algorithm offers new insights into the molecular basis of diseases.
• CSMR holds significant potential for advancing personalized medicine by clarifying genetic-phenotype relationships.