A decision-theoretical alternative to testing many hypotheses

Area of Science:

• Bioinformatics
• Statistical genomics
• Computational biology

Background:

• Multiple hypothesis testing is crucial in high-throughput studies like microarray analysis.
• Existing methods often focus on controlling the false discovery rate (FDR).
• Failure to discover (false negative) is also a critical consideration often overlooked.

Purpose of the Study:

• To develop a decision-theoretical framework for multiple hypothesis testing.
• To incorporate uneven losses for two types of errors (false discovery and failure to discover).
• To minimize the total expected loss across all classifications.

Main Methods:

• A decision-theoretical approach was formulated.
• Uneven loss functions were assigned to misclassifications.
• Optimization was performed to minimize total expected loss for a collection of decisions.

Main Results:

• The proposed method provides a principled way to balance false discoveries and failures to discover.
• It accounts for differential costs associated with different types of errors.
• Minimizing total expected loss offers a robust strategy for hypothesis testing in complex experiments.

Conclusions:

• The decision-theoretical approach offers an improved method for multiple hypothesis testing.
• Considering uneven error losses enhances the identification of true exceptions in large-scale studies.
• This framework is applicable to microarray analysis and other fields with extensive hypothesis testing.