SYBA

SYBA classifies organic compounds as easy-to-synthesize (ES) or hard-to-synthesize (HS) using a fragment-based Bernoulli naïve Bayes model to assess synthetic accessibility.

Key Features:

• Fragment-Based Classification: Infers synthetic accessibility from constituent molecular fragments by attributing fragment-level contributions.
• Bernoulli naïve Bayes Model: Uses a Bernoulli naïve Bayes classifier to model fragment presence/absence for ES versus HS classification.
• SYBA Scores: Computes SYBA scores by assigning contributions to individual fragments based on their observed frequencies in ES and HS sets.
• Training and Databases: Trained on easy-to-synthesize molecules from ZINC15 and hard-to-synthesize molecules generated via Nonpher.
• Comparative Performance: Evaluated against a random forest baseline and the synthetic accessibility methods SAScore and SCScore, with improvements over SAScore and SCScore at their suggested thresholds and marginal improvement over the random forest baseline.
• Threshold Optimization: Notes that adjusting the SAScore threshold (from 6.0 to -4.5) can produce performance comparable to SYBA, indicating parameter sensitivity.
• Analytical Utility: Enables dissection of individual fragment contributions to identify molecular parts that influence synthetic accessibility.
• Implementation: Implemented as a Python-based bioinformatics package.

Scientific Applications:

• Medicinal Chemistry: Prioritizing and evaluating the synthetic accessibility of candidate molecules in medicinal chemistry workflows.
• Drug Discovery: Assisting compound selection for synthesis in drug discovery to improve screening-to-synthesis efficiency.
• Design and Route Optimization: Providing fragment-level insights to guide compound design and optimization of synthetic routes.

Methodology:

SYBA assigns fragment contributions based on observed fragment frequencies in ES molecules from ZINC15 and HS molecules generated by Nonpher, and applies a Bernoulli naïve Bayes classifier on fragment presence/absence to compute compound-level SYBA scores.