StructuralDPPIV: a novel deep learning model based on atom structure for predicting dipeptidyl peptidase-IV...

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Diabetes mellitus is a global health crisis, leading to severe complications like blindness and cardiovascular disease.
• Dipeptidyl peptidase IV inhibitory peptides (DPP-IV-IPs) represent a promising class of anti-diabetic therapeutics.
• Current peptide-mining tools hinder the efficient discovery of novel DPP-IV-IPs.

Purpose of the Study:

• To introduce StructuralDPPIV, a novel deep learning model for identifying DPP-IV-IPs.
• To enhance the discovery pipeline for next-generation anti-diabetic drugs.

Main Methods:

• Developed StructuralDPPIV, a deep learning model integrating molecular graph features and amino acid sequence information.
• Validated the model's performance on independent test and wet experiment datasets.
• Employed Class Activation Mapping (CAM) and perturbation experiments for model interpretability.

Main Results:

• StructuralDPPIV demonstrated superior performance compared to existing state-of-the-art methods.
• The model achieved high accuracy in identifying DPP-IV-IPs.
• Interpretability analyses provided insights into the model's predictive reasoning.

Conclusions:

• StructuralDPPIV is an effective and interpretable tool for DPP-IV-IP identification.
• The model accelerates the discovery of potential anti-diabetic drugs.
• The open-source availability of the code facilitates further research and development.