Transfer learning for accelerated failure time model with microarray data.

Area of Science:

• Bioinformatics
• Genomics
• Statistical Modeling

Background:

• Microarray studies aim to identify genes linked to disease progression.
• Limited sample sizes in rare diseases hinder accurate gene selection and risk assessment.
• Leveraging external data (source cohorts) is crucial for improving analysis of target cohorts.

Purpose of the Study:

• To develop a transfer learning method for the Accelerated Failure Time (AFT) model.
• To enhance gene selection and risk prediction in target cohorts using information from source cohorts.
• To address challenges posed by limited sample sizes and cohort heterogeneity.

Main Methods:

• Proposed a transfer learning approach for the AFT model.
• Employed adaptive information borrowing from source cohorts to improve target cohort fitting.
• Utilized Leave-One-Out cross-validation to assess gene stability and predictive power.

Main Results:

• Transfer learning method accurately identifies key genes with reduced estimation error compared to methods without external data.
• The approach demonstrates robustness in handling cross-cohort heterogeneity.
• Analysis of GSE88770 and GSE25055 data showed stable gene selection and satisfactory risk prediction.

Conclusions:

• The proposed transfer learning method effectively improves gene identification and risk prediction in microarray studies with limited samples.
• It offers a robust solution for analyzing heterogeneous cohort data.
• The method provides a valuable tool for advancing prognostic research in genomics.