A Novel Feature Selection Method for Uncertain Features: An Application to the Prediction of Pro-/Anti-Longevity...

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Understanding the ageing process is a complex biological challenge.
• Machine learning classification methods are increasingly used to predict genes influencing ageing or longevity.
• Protein-Protein Interaction (PPI) features are valuable but often contain uncertainty (confidence scores) ignored by traditional methods.

Purpose of the Study:

• To develop a novel feature selection method that effectively handles uncertainty in PPI confidence scores.
• To enhance the accuracy of gene classification models for ageing and longevity prediction.
• To improve the interpretability of classification results by reducing the number of selected features.

Main Methods:

• Proposed the Lazy Feature Selection for Uncertain Features (LFSUF) method.
• LFSUF employs a lazy learning paradigm, selecting features per instance for flexible classification.
• The method explicitly addresses uncertainty in PPI confidence scores.

Main Results:

• LFSUF achieved superior predictive accuracy compared to methods that ignore PPI confidence scores or handle uncertainty globally.
• The method demonstrated improved performance by using a per-instance feature selection approach.
• Interpretation of results showed a significant reduction in selected features, enhancing interpretability.

Conclusions:

• LFSUF offers an effective approach for feature selection in the presence of uncertain PPI data.
• The method advances machine learning applications in ageing and longevity research.
• The per-instance feature selection strategy provides a more flexible and accurate classification framework.