Generic comparison of protein inference engines

Area of Science:

• Proteomics
• Biochemistry
• Computational Biology

Background:

• Shotgun proteomics relies on protein identifications, but objective assessment of inference approaches is lacking.
• Debate continues regarding the best methods for inferring and reporting protein identifications.

Purpose of the Study:

• To introduce a formal performance measure for objectively assessing protein inference strategies.
• To explore the impact of excluding unreliable protein identifications, like single-hit wonders.

Main Methods:

• Developed an intuitive, generic performance measure for protein inference.
• Created a family of protein inference engines by varying exclusion criteria.
• Benchmarked engines on diverse proteomic datasets and mass spectrometry platforms.

Main Results:

• Optimally performing inference engines retained all high-confidence spectral evidence.
• Exclusion of specific identification types, such as single-hit wonders, was not consistently beneficial.
• Performance measure enabled selection of optimal strategies for specific datasets.

Conclusions:

• No single rigid rule for protein inference suits all datasets.
• Advocates for using the performance measure to assess multiple approaches for maximal reliable proteome coverage.
• Experimentalists should tailor protein inference strategies to their specific data context.