False discovery rate regression: an application to neural synchrony detection in primary visual cortex.

Area of Science:

• Statistics
• Computational Neuroscience
• Bioinformatics

Background:

• Traditional multiple testing methods assume a global false-discovery rate (FDR) analysis, which may be unsuitable for large-scale screening.
• Combined analyses can result in poorly calibrated error rates across different experimental subsets when auxiliary information is available.

Purpose of the Study:

• Introduce false-discovery-rate (FDR) regression, a novel approach that directly incorporates auxiliary information into multiple testing.
• Address limitations of global FDR analyses in large-scale screening by utilizing test-specific covariates.

Main Methods:

• Developed FDR regression, a method motivated by a two-groups model where covariates influence the local false discovery rate.
• Investigated computational and inferential challenges associated with FDR regression variations.
• Applied the method to analyze neural recordings from the primary visual cortex to detect fine-time-scale neuronal interactions.

Main Results:

• FDR regression significantly improves statistical power for a fixed false-discovery rate when covariate effects are present.
• The method demonstrates robustness, avoiding inflated error rates when covariate effects are absent.
• Detected approximately 50% more synchronous neuronal pairs compared to standard FDR-controlling analyses in neural recordings.

Conclusions:

• FDR regression offers a powerful and robust alternative to traditional multiple testing methods, especially in large-scale screening scenarios with available auxiliary data.
• The approach enhances the detection of significant findings, as demonstrated by its application in identifying neuronal synchrony.
• An R package, FDRreg, is available to implement the described FDR regression methods.