SCScore: Synthetic Complexity Learned from a Reaction Corpus.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Organic Synthesis

Background:

• Existing definitions of molecular complexity aid in prioritizing drug leads and designing synthetic routes.
• A quantitative measure of synthetic complexity is needed to optimize chemical synthesis and retrosynthetic analysis.

Purpose of the Study:

• To reformalize the definition of synthetic complexity to correlate with the number of reaction steps.
• To develop a predictive metric for synthetic complexity using machine learning.

Main Methods:

• A neural network model was trained on 12 million reactions from the Reaxys database.
• The model learned a pairwise inequality constraint: reaction products are generally more complex than reactants.
• The resulting metric is termed the Synthetic Complexity Score (SCScore).

Main Results:

• The SCScore demonstrates nonlinear behavior, effectively capturing increases in complexity along synthetic pathways.
• The metric aligns with the principle that reaction products are synthetically more complex than starting materials.
• The SCScore shows promise for prioritizing synthetic targets and evaluating reaction efficiency.

Conclusions:

• The developed SCScore provides a data-driven approach to quantify synthetic complexity.
• This metric can guide retrosynthetic planning and the discovery of complex molecules.
• The SCScore offers a valuable tool for optimizing chemical synthesis and advancing drug discovery.