GASA

GASA predicts the synthetic accessibility of small molecules using a graph attention–based approach to inform compound prioritization in drug discovery.

Key Features:

• Graph Neural Network Architecture: Utilizes a graph neural network that processes molecular graphs to capture complex structural relationships within molecules.
• Attention Mechanism: Applies a graph attention mechanism to identify and prioritize structural features that influence synthetic accessibility.
• Improved Distinguishability: Employs sampling around hypothetical classification boundaries to enhance differentiation between structurally similar molecules.
• Extensive Evaluation and Comparison: Evaluated against descriptor-based machine learning methods such as random forest and eXtreme gradient boosting and compared to synthetic accessibility scores SYBA, SCScore, RAscore, and SAscore, demonstrating superior performance and a broader applicability domain.
• Feature Importance Analysis: Assigns attention weights to individual atoms to provide interpretable insights into molecular features affecting synthetic accessibility.

Scientific Applications:

• Drug discovery prioritization: Predicts ease-of-synthesis to inform selection and resource allocation for potential drug candidates.
• Compound screening and triage: Aids in prioritizing compounds that are more feasible to synthesize, thereby streamlining the drug development pipeline.

Methodology:

Construct molecular graphs from chemical structures and apply a graph attention–based GNN that performs self-feature deduction and incorporates sampling around classification boundaries to refine synthetic accessibility predictions.