Regression-based heterogeneity analysis to identify overlapping subgroup structure in high-dimensional data.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Complex diseases exhibit significant heterogeneity, making analysis challenging.
• Existing regression-based heterogeneity analyses often assume disjoint subgroups, limiting their applicability.
• Real-world biological data frequently involves overlapping structures, such as genes with multiple functions.

Purpose of the Study:

• To develop a novel regression-based heterogeneity analysis method that accommodates overlapping subgroups.
• To address high-dimensional data and small sample size limitations in heterogeneity analysis.
• To improve the understanding of disease biology by identifying complex subgroup structures.

Main Methods:

• Introduced a subgroup membership vector for each sample to allow for overlapping memberships.
• Developed a novel loss function incorporating an L0 norm penalty for membership vectors to handle small sample sizes.
• Applied sparse penalization for regularized estimation and feature selection in high-dimensional data.

Main Results:

• The proposed method demonstrated superior performance compared to existing approaches in extensive simulations.
• Analysis of Cancer Cell Line Encyclopedia and The Cancer Genome Atlas lung cancer data revealed overlapping subgroup structures.
• The approach showed favorable performance in prediction and stability on real-world biological datasets.

Conclusions:

• The novel approach effectively identifies overlapping subgroup structures in complex diseases.
• This method provides a more realistic and powerful tool for heterogeneity analysis in biology and medicine.
• The findings suggest improved disease subtyping and personalized treatment strategies.