fdrci: FDR confidence interval selection and adjustment for large-scale hypothesis testing

Area of Science:

• Biostatistics
• Genomics
• Bioinformatics

Background:

• Traditional fixed discovery thresholds (e.g., 0.05) can prevent the identification and publication of weak but potentially real biological effects.
• Existing false discovery rate (FDR) estimation methods lack robust strategies for selecting discovery thresholds post hoc.

Purpose of the Study:

• To develop and present a flexible statistical approach for estimating FDR.
• To enable principled post hoc selection of discovery thresholds beyond the conventional 0.05.
• To facilitate the identification of true discoveries among potentially weak effects.

Main Methods:

• Utilized a permutation-based FDR estimator for enhanced precision.
• Proposed a series of discovery thresholds.
• Employed an FDR confidence interval selection and adjustment technique to identify intervals not containing one, indicating true discoveries.

Main Results:

• Successfully applied the method to a transcriptome-wide association study (TWAS) within the MAVERICC clinical trial for metastatic colorectal cancer.
• Identified several genes where predicted expression is significantly associated with progression-free or overall survival.
• Demonstrated the utility of the flexible FDR approach in a real-world clinical genomics dataset.

Conclusions:

• The proposed flexible FDR estimation method allows for more nuanced threshold selection, improving the discovery of subtle effects.
• This approach enhances the ability of researchers to identify significant biological signals in complex datasets like TWAS.
• Available R software (fdrci package on CRAN) facilitates the implementation of this advanced statistical technique.