Extracting high confidence protein interactions from affinity purification data: at the crossroads.

Area of Science:

• Proteomics
• Biochemistry
• Bioinformatics

Background:

• Affinity-purification and mass spectrometry (AP-MS) are key techniques for identifying protein-protein interactions (PPIs).
• Scoring methods are crucial for assessing the reliability of putative PPIs derived from AP-MS data.
• Selecting the appropriate scoring method for AP-MS datasets presents a significant challenge.

Purpose of the Study:

• To compare the performance of six popular scoring methods for AP-MS data.
• To evaluate the consistency and reliability of high-confidence PPI networks generated by these methods.
• To identify limitations of current scoring methods in handling noisy AP-MS data.

Main Methods:

• Application of six distinct scoring methods to six diverse AP-MS datasets (human, fly, yeast).
• Comparative analysis of scoring method performance across different biological contexts and proteome coverage.
• Evaluation of overlap and properties of high-confidence PPI networks generated by each method.

Main Results:

• Scoring method performance varied substantially depending on the specific AP-MS dataset.
• High-confidence PPI networks generated by different methods showed very poor overlap (1.7-4.1% common interactions).
• Current scoring methods primarily remove obvious contaminants but fail to reliably identify specific interactions from spurious associations.

Conclusions:

• Existing scoring methods do not perform as expected due to high noise levels in AP-MS data and their limited capacity to handle it.
• Biases in benchmarking using Gold Standard datasets further complicate method evaluation.
• Addressing noise in raw data, improving scoring method robustness, and refining benchmarking strategies are essential for advancing PPI research.