Ensemble methods for testing a global null.

Area of Science:

• Statistics
• Bioinformatics
• Machine Learning

Background:

• Global null hypothesis testing is crucial across various applications.
• Developing powerful tests for specific alternative classes remains challenging.
• Existing methods often rely on prior knowledge to improve test power.

Purpose of the Study:

• To propose a novel ensemble testing framework for robust global null hypothesis testing.
• To leverage ensemble learning principles, similar to random forests, for improved statistical power.
• To address challenges in designing tests powerful against specific alternative classes.

Main Methods:

• Developed an ensemble framework aggregating multiple weak base tests.
• Applied the framework to four global testing problems in Whole Genome Sequencing (WGS) association studies.
• Established theoretical optimality using Bahadur efficiency.

Main Results:

• The proposed ensemble tests demonstrated strong and robust power for global nulls.
• Simulations confirmed type I error control and power gains.
• Analysis of a real WGS dataset validated the framework's practical utility.

Conclusions:

• The ensemble testing framework offers a powerful and robust approach to global null hypothesis testing.
• This method shows significant promise for applications in genetic association studies.
• The framework provides a flexible strategy for enhancing statistical power in complex testing scenarios.