Automatic peak selection by a Benjamini-Hochberg-based algorithm.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Computational methods in bioinformatics generate numerous predictions requiring selection for accuracy.
• Selecting the optimal number of predictions is crucial for maintaining high precision and capturing true positives.
• In NMR-based protein structure determination, manual selection of peaks from thousands of candidates is common but time-consuming.

Purpose of the Study:

• To develop an automated method for selecting candidate peaks in bioinformatics.
• To address the challenge of determining the optimal number of predictions to maximize true positives while maintaining precision.
• To improve peak selection in NMR-based protein structure determination.

Main Methods:

• Formulated peak selection as a multiple testing problem.
• Developed a Benjamini-Hochberg (B-H)-based approach to automatically select peaks.
• Converted candidate peaks into p-values and applied the B-H algorithm for selection.

Main Results:

• The B-H-based approach significantly increased the number of true peaks identified compared to traditional fixed-number methods.
• Reduced missing peak rates by an average of 20% (WaVPeak) and 26% (PICKY) in benchmark datasets.
• Achieved 88% recall and 83% precision using a consensus of B-H-selected peaks from two methods.

Conclusions:

• The proposed B-H-based method offers an automated and effective solution for peak selection in bioinformatics.
• This approach enhances accuracy and efficiency in NMR-based protein structure determination.
• The method is adaptable for other prediction selection problems in bioinformatics and is publicly available.