Robust Transfer Learning for High-Dimensional GLM Using γ $$ \gamma $$ -Divergence With Applications to Cancer...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional profiling data is crucial for complex disease analysis, risk assessment, and biomarker discovery.
• Limited sample sizes in cancer genomics studies necessitate information borrowing from external data sources.
• Existing transfer learning methods lack robustness to outliers and data contamination common in biomedical data.

Purpose of the Study:

• To develop a robust transfer learning approach for high-dimensional genomic data analysis.
• To address the limitations of existing methods regarding data outliers and contamination.
• To improve estimation and prediction accuracy in complex disease studies.

Main Methods:

• Proposed a robust transfer learning method using minimum γ-divergence within a generalized linear model (GLM) framework.
• Incorporated a data-driven source detection scheme to identify informative sources and prevent negative transfer.
• Developed a computationally efficient algorithm based on proximal gradient descent for transfer and debiasing.

Main Results:

• Established theoretical guarantees including consistency and high-dimensional estimation error bounds.
• Demonstrated superior and competitive performance in selection, prediction, and classification via simulations.
• Validated practical utility on real-world breast cancer and glioblastoma genomic data.

Conclusions:

• The proposed robust transfer learning method enhances the analysis of high-dimensional genomic data in complex diseases.
• The approach effectively handles data imperfections, offering reliable performance and improved accuracy.
• It holds significant potential for advancing biomarker discovery and risk assessment in cancer research.