NeSyDPP4-QSAR: A Neuro-Symbolic AI Approach for Potent DPP-4-Inhibitor Discovery in Diabetes Treatment

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Bioinformatics and computational biology

Background:

• Diabetes Mellitus (DM) is a global health crisis, with Dipeptidyl Peptidase-4 (DPP-4) being a key therapeutic target for type 2 diabetes.
• Existing DPP-4 inhibitors have limitations, including adverse effects, necessitating the search for novel therapeutic agents.
• In silico methods like Quantitative Structure-Activity Relationship (QSAR) modeling offer efficient alternatives to experimental drug screening.

Purpose of the Study:

• To develop and evaluate a novel computational model for predicting DPP-4 inhibitors.
• To compare the performance of a neuro-symbolic approach against deep neural networks and transformers for QSAR modeling.
• To identify accurate in silico methods for classifying biological activities and discovering novel anti-diabetic drugs.

Main Methods:

• A dataset of 6,563 DPP-4 bioactivity records was compiled from public databases.
• A neuro-symbolic model (NeSyDPP4-QSAR) was developed using diverse features, including descriptors, fingerprints, and chemical language model embeddings.
• The model's performance was benchmarked against deep neural network (DNN) and transformer baseline models using metrics like accuracy, F1-score, ROC AUC, and MCC.

Main Results:

• The NeSyDPP4-QSAR model achieved high predictive accuracy (0.9725), F1-score (0.9723), ROC AUC (0.9719), and MCC (0.9446).
• The neuro-symbolic approach outperformed standard DNN and transformer models in predicting DPP-4 inhibition.
• Key features contributing to the model's success included CDK Extended and Morgan fingerprints.

Conclusions:

• The integration of neuro-symbolic strategies offers significant potential for advancing drug discovery in diabetes.
• The NeSyDPP4-QSAR model provides a robust and accurate platform for identifying novel DPP-4 inhibitors.
• This computational approach can accelerate the classification of biological activities and the development of new anti-diabetic medications.