Added predictive value of omics data: specific issues related to validation illustrated by two case studies

Area of Science:

• Biostatistics
• Bioinformatics
• Genomics

Background:

• Independent validation of gene signatures is essential.
• Focus has shifted to validating the added predictive value of gene signatures.
• Gene signatures are increasingly integrated with clinical predictors.

Purpose of the Study:

• To demonstrate methods for validating the added predictive value of gene signatures derived from high-dimensional data.
• To critically assess the impact of methodological choices on validation results.
• To compare different validation techniques in a high-dimensional setting.

Main Methods:

• Developing and applying validation techniques for gene signatures in high-dimensional data.
• Utilizing repeated cross-validation procedures.
• Analyzing the added predictive value of two gene signatures in leukemia patient survival studies.

Main Results:

• High-dimensional omics data presents unique challenges for validation.
• Different validation techniques yield varying results.
• Empirical comparison of techniques illustrates their performance.

Conclusions:

• The validation of gene signatures is influenced by the high-dimensional nature of omics predictors.
• A robust analysis procedure based on repeated cross-validation is recommended.
• Ensuring the added predictive value of gene signatures is key for reliable clinical application.