An explainable stacking-based approach for accelerating the prediction of antidiabetic peptides

Area of Science:

• Computational biology
• Biotechnology
• Bioinformatics

Background:

• Diabetes mellitus is a chronic condition marked by hyperglycemia, necessitating effective management strategies.
• Current experimental therapies require robust methods for efficacy evaluation.
• Computational tools offer a promising avenue for accelerating the discovery of novel diabetes treatments.

Purpose of the Study:

• To introduce STADIP, a novel stacking-based ensemble predictor for identifying antidiabetic peptides (ADPs).
• To leverage machine learning for enhanced prediction of therapeutic peptides in diabetes management.

Main Methods:

• Developed 84 baseline models using 12 feature encodings and 7 machine learning techniques.
• Employed a two-step feature selection (XGB-SFS) to optimize predictive performance.
• Integrated 45 selected features into a final hybrid model using a meta-predictor approach (XGB classifier).

Main Results:

• STADIP demonstrated superior predictive performance over individual baseline models.
• Achieved high accuracy (0.954) and Matthew's correlation coefficient (0.877) in independent tests.
• The stacked ensemble approach effectively enhanced ADP prediction accuracy.

Conclusions:

• STADIP is a robust and accurate tool for predicting antidiabetic peptides.
• This computational approach can significantly aid researchers in discovering new therapeutic peptides for diabetes.
• The study highlights the potential of machine learning in advancing diabetes treatment research.