Measuring the wisdom of the crowds in network-based gene function inference.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Network-based gene function inference methods are widely used but progress is unclear.
• Controlling for data and algorithm implementation is crucial for performance evaluation.

Purpose of the Study:

• To evaluate the performance trends of gene function inference methods.
• To assess the impact of data and algorithm aggregation on prediction accuracy.

Main Methods:

• Utilized well-characterized algorithms to generate 'untweaked' results.
• Controlled for underlying biological network data across different tests.
• Measured performance using standard metrics like area under the ROC curve (AUROC).

Main Results:

• State-of-the-art machine learning methods achieve 'gold standard' performance.
• Algorithm performance is highly aligned when controlling for data.
• Algorithm aggregation offers modest benefits (17% AUROC increase).
• Data aggregation yields substantial gains (88% AUROC improvement).

Conclusions:

• Additional algorithm development offers little improvement for gene function prediction.
• Data aggregation is a more effective strategy for enhancing prediction accuracy.