Modeling Between-Study Heterogeneity for Improved Replicability in Gene Signature Selection and Clinical Prediction

Area of Science:

• Genomics
• Biostatistics
• Translational Medicine

Background:

• Gene signatures are crucial for predicting disease outcomes and guiding clinical decisions.
• Replicability issues in gene signature identification limit their clinical applicability and generalizability.
• Between-study heterogeneity poses a significant challenge in integrating genomic data.

Purpose of the Study:

• To develop a novel, robust method for selecting reproducible gene signatures from multiple genomic datasets.
• To address and account for between-study heterogeneity in gene signature identification.
• To improve the clinical utility and generalizability of gene signatures for disease subtyping and outcome prediction.

Main Methods:

• A novel approach integrating rank-based quantities for selecting gene signatures with consistently non-zero effects.
• Utilizing a high-dimensional penalized Generalized Linear Mixed Model (pGLMM) to handle data heterogeneity.
• Comparing the proposed method against traditional strategies that disregard between-study heterogeneity.

Main Results:

• Asymptotic results theoretically justify the method's performance.
• Simulation studies demonstrate the method's advantage in handling data heterogeneity.
• The approach was successfully applied to subtype pancreatic cancer patients using four gene expression studies.

Conclusions:

• The proposed method enhances the replicability and reliability of gene signatures.
• Accounting for between-study heterogeneity is critical for developing clinically applicable gene signatures.
• This approach offers a promising tool for advancing personalized medicine through robust genomic data analysis.