Prioritizing hypothesis tests for high throughput data.

Area of Science:

• Genomics
• Statistical Bioinformatics
• High-throughput Data Analysis

Background:

• High-throughput studies generate massive data, increasing hypothesis tests and necessitating stringent significance thresholds.
• Filtering methods use independent information to reduce multiple testing and enhance discovery, but their optimal application remains unclear.

Purpose of the Study:

• To quantify the impact of filter information quality and cutoff selection on the effectiveness of filtering methods.
• To develop a data-based approach for selecting filter cutoffs that maintains error control and maximizes discovery probability.

Main Methods:

• Quantified the relationship between filter quality, redundancy, cutoff choice, and discovery probability.
• Introduced a data-driven method for cutoff selection with a correction factor for family-wise error rate control.
• Developed a P-value weighting method to further enhance performance.

Main Results:

• Filtering significantly increases discovery probability (up to 10-fold) with high-quality, redundant information.
• Effectiveness diminishes with lower quality or less redundant information.
• A data-based cutoff selection method provides several-fold discovery advantage while controlling Type I error.

Conclusions:

• The quality and redundancy of filter information are critical for effective filtering.
• A data-based cutoff selection strategy is essential for maximizing discoveries and controlling errors in high-throughput biological studies.
• The proposed methods offer substantial improvements over traditional filtering approaches.