Integrating data augmentation and BERT-based deep learning for predicting alpha-glucosidase inhibitors derived from...

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Diabetes mellitus is a global health concern exacerbated by obesity and inactivity.
• Alpha-glucosidase inhibitors are sought as therapeutic agents for diabetes management.
• Natural products offer a promising avenue for discovering novel enzyme inhibitors.

Purpose of the Study:

• To identify potent alpha-glucosidase inhibitors using deep learning and data augmentation.
• To evaluate the efficacy of augmented deep learning models in drug discovery.
• To investigate actaeaepoxide 3-O-xyloside as a potential anti-diabetic compound.

Main Methods:

• Generated diverse SMILES strings and applied data augmentation techniques.
• Fine-tuned pre-trained deep learning models (e.g., PC10M-450k) from Hugging Face.
• Performed molecular docking and molecular dynamics (MD) simulations for stability analysis.

Main Results:

• The PC10M-450k model demonstrated superior performance in identifying inhibitors.
• Actaeaepoxide 3-O-xyloside was identified as a potent alpha-glucosidase inhibitor.
• Simulations confirmed stable enzyme-compound interaction and high inhibition probability compared to acarbose.

Conclusions:

• Actaeaepoxide 3-O-xyloside is a promising candidate for diabetes therapy.
• Data augmentation and pre-trained models significantly accelerate in silico drug discovery.
• This approach offers a pathway to develop more effective therapeutic solutions for diabetes.