Discovery of potential antidiabetic peptides using deep learning.

Area of Science:

• Biochemistry and bioinformatics
• Computational drug discovery
• Peptide science

Background:

• Antidiabetic peptides (ADPs) are crucial for diabetes management but their discovery is hindered by data limitations and experimental costs.
• Traditional methods for identifying ADPs are time-consuming and expensive, necessitating advanced computational approaches.

Purpose of the Study:

• To develop and evaluate deep learning models for accurate and efficient prediction of antidiabetic peptides.
• To address the challenges in ADP discovery by leveraging advanced computational techniques.
• To explore the generation and screening of novel ADPs for therapeutic potential.

Main Methods:

• Development of two deep learning models: a single-channel CNN and a CNN+RNN+Bi-LSTM network.
• Data preprocessing using the evolutionary scale model (ESM-2) and 10-fold cross-validation for model training and evaluation.
• Generation of new candidate ADPs using SeqGAN and screening with the CNN model, followed by physicochemical and structural property evaluation.

Main Results:

• The single-channel CNN model demonstrated superior performance, achieving 90.48% accuracy on an independent test set of newly identified ADPs.
• The developed models surpassed existing tools for antidiabetic peptide prediction.
• The study successfully screened potential ADPs with favorable physicochemical and structural properties for pharmaceutical applications.

Conclusions:

• The study established robust deep learning models for predicting antidiabetic peptides, significantly advancing the field.
• The developed models can be effectively applied to discover and screen novel peptide candidates for antidiabetic therapies.
• This research addresses a critical need for efficient methods in peptide-based antidiabetic drug discovery.