Robust Distribution-Free Tests for the Linear Model.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Biological data frequently exhibits heavy-tailed and skewed noise, complicating statistical analysis.
• Existing methods struggle to maintain accuracy and control error rates under such conditions.

Purpose of the Study:

• To introduce RobustPALMRT, a flexible permutation framework for robustly testing covariate associations in biological data.
• To enhance statistical power and type I error control in the presence of heavy-tailed or skewed noise.

Main Methods:

• Developed RobustPALMRT, a permutation framework applicable to robust and quantile regressions.
• Incorporated a robust loss function in the model-evaluation step, separating it from model-fitting.
• Introduced DispersionPALMRT for detecting differences in dispersion between groups.

Main Results:

• RobustPALMRT controls type I error rates for finite samples, even with heavy-tailed or skewed noise.
• The framework demonstrates improved performance by separating model-fitting and evaluation, particularly with robust loss functions.
• Analysis of Long-COVID immunological data revealed novel differences using RobustPALMRT, highlighting its utility in complex biological datasets.

Conclusions:

• RobustPALMRT provides a powerful and flexible tool for analyzing biological data with challenging noise characteristics.
• The method enhances the detection of subtle associations and group differences, advancing biological data interpretation.
• This framework has significant implications for fields like immunology and personalized medicine, especially when dealing with noisy, real-world data.