AntiDMPpred

AntiDMPred predicts whether peptide sequences are anti-diabetic bioactive peptides to facilitate identification of candidate therapeutic peptides for diabetes mellitus (DM).

Key Features:

• Dataset: A benchmark dataset comprising 236 anti-diabetic peptides and 236 non-anti-diabetic peptides was curated for model development.
• Feature representation: Peptide sequences were encoded using four types of sequence-derived descriptors.
• Feature selection: Non-redundant features were selected using a combination of four machine learning methods and six feature scoring techniques.
• Model training: Selected features were used to train multiple machine learning classifiers, with a random forest classifier demonstrating superior performance.
• Evaluation: Nested five-fold cross-validation reported an accuracy of 77.12% and an area under the receiver operating characteristic curve (AUCROC) of 0.8193.

Scientific Applications:

• Anti-diabetic peptide discovery: Prioritizes candidate peptides for experimental validation in peptide-based drug development for diabetes mellitus.
• Screening and prioritization: Enables selection of promising peptide leads to focus experimental resources and accelerate follow-up studies.

Methodology:

Model development used a benchmark dataset (236 positive, 236 negative), four sequence-derived descriptor types, feature selection via four machine learning methods and six feature scoring techniques, training of diverse classifiers including random forest, and evaluation by nested five-fold cross-validation.