Multimodal Bonds Reconstruction Towards Generative Molecular Design.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Molecular modeling

Background:

• Generative models accelerate de novo drug design by creating novel 2D and 3D molecular structures.
• Accurate chemical bond reconstruction, particularly from distorted geometries, is a significant challenge in generative chemistry.

Purpose of the Study:

• To introduce YuelBond, a multimodal graph neural network framework for robust chemical bond reconstruction.
• To address bond reconstruction from accurate 3D coordinates, perturbed geometries of de novo generated compounds, and 2D topological graphs.

Main Methods:

• Development of YuelBond, a multimodal graph neural network.
• Evaluation of YuelBond's performance on standard 3D structures and crude de novo generated compounds (CDGs).
• Comparison with traditional rule-based methods like RDKit.

Main Results:

• YuelBond achieved a 98.4% F1-score on standard 3D structures.
• YuelBond demonstrated strong performance with a 92.7% F1-score on distorted CDGs, outperforming RDKit.
• The framework successfully reconstructs bonds from imperfect molecular data.

Conclusions:

• YuelBond provides accurate and reliable bond reconstruction, addressing a critical gap in generative drug discovery.
• The method enhances the utility of generative models by enabling robust handling of imperfect molecular geometries.
• YuelBond represents a significant advancement for computational drug design pipelines.