Correct machine learning on protein sequences: a peer-reviewing perspective.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Machine learning (ML) is increasingly used for predicting protein features from sequences.
• ML in bioinformatics offers power but risks introducing biases, potentially overestimating performance.

Purpose of the Study:

• To provide guidelines for authors and peer reviewers to avoid common pitfalls in ML applications.
• To ensure robust and reliable performance evaluation of ML methods in bioinformatics.

Main Methods:

• Emphasize the necessity of biological understanding for creating large, diverse datasets.
• Stress the importance of separating training and testing processes to prevent performance overestimation.
• Advocate for comparing novel predictors against existing methods and baseline strategies.

Main Results:

• Adherence to guidelines helps mitigate unexpected biases in ML models.
• Proper validation prevents over-selling the performance of new prediction methods.
• Ensures that ML applications in bioinformatics are both useful and reliable.

Conclusions:

• Implementing these guidelines enhances the rigor of ML research in bioinformatics.
• Facilitates accurate assessment of ML model performance for protein feature prediction.
• Aids non-specialists in understanding critical issues in bioinformatics ML.