Generating novel molecule for target protein (SARS-CoV-2) using drug-target interaction based on graph neural network

Area of Science:

• Computational chemistry and drug discovery
• Artificial intelligence in medicine
• Virology and molecular biology

Background:

• The global spread of SARS-CoV-2 necessitates the development of effective antiviral treatments.
• Existing machine learning methods for drug discovery often overlook the 3D structural properties of molecules and proteins, leading to limitations in predicting binding efficacy.
• A significant portion of potential drug candidates are computationally discarded due to poor binding affinity.

Purpose of the Study:

• To develop a computational framework for generating highly potent small molecules capable of binding to SARS-CoV-2 protein structures.
• To address the limitations of 1D sequence representations by incorporating 3D structural information of molecules and proteins.
• To improve the efficiency and accuracy of drug discovery for novel antiviral agents.

Main Methods:

• A novel framework integrating Gated Graph Neural Networks (GGNN), Knowledge Graphs, and an Early Fusion approach.
• Utilized both tertiary (3D) and sequential representations for molecules and protein targets.
• Employed Knowledge Graphs for pre-screening generated molecules, significantly reducing the search space before applying the Early Fusion model for binding affinity prediction.

Main Results:

• The framework successfully generated valid, unique, and chemically accurate molecules with high precision.
• Knowledge Graph screening reduced the molecular dataset by approximately 96%, while retaining over 85% of desirable binding molecules and rejecting over 99% of ineffective ones.
• The model demonstrated high accuracy in predicting binding affinity scores for potential drug candidates against SARS-CoV-2 targets like RNA-dependent-RNA polymerase (RdRp) and 3C-like protease (3CLpro).

Conclusions:

• The proposed framework effectively leverages 3D structural data and advanced machine learning techniques for efficient and accurate SARS-CoV-2 drug discovery.
• Incorporating Knowledge Graphs significantly enhances the screening process, minimizing computational resources and time.
• This approach holds promise for accelerating the development of novel antiviral therapies against emerging infectious diseases.