The State of Peptide Detectability in Computational Proteomics and Guidelines for AI Applications

Area of Science:

• Computational Biology
• Proteomics
• Artificial Intelligence

Background:

• Artificial intelligence (AI) has significantly advanced proteomics, particularly in mass spectrometry tasks like peptide detectability prediction.
• The increasing volume of proteomics data necessitates robust, reproducible, and comparable AI applications.

Purpose of the Study:

• To conduct a comprehensive scoping review of AI techniques for peptide detectability prediction.
• To identify challenges and propose guidelines for improving AI methods in computational proteomics.

Main Methods:

• Analysis of over 25 peer-reviewed methods for peptide detectability published between 2006 and 2025.
• Evaluation of algorithmic sophistication versus adherence to machine learning standards.

Main Results:

• A persistent discrepancy exists between advanced AI algorithms and consistent application of machine learning standards.
• Common issues include heterogeneous dataset construction, lack of transparency in model design/evaluation, and limited reproducibility.
• Recent tools show that adherence to rigorous standards is achievable.

Conclusions:

• Actionable guidelines are proposed for transparent reporting, dataset separation, performance evaluation, and reproducibility in AI for proteomics.
• Future research should focus on integrating structural biology, using data-independent acquisition datasets, and developing explainable AI for interpretability and transferability.