An Expert-Augmented Deep Learning Approach for Synthesis Route Evaluation

Selection of efficient multi-step synthesis routes is a fundamental challenge in organic synthesis. Comparing different routes involves numerous parameters, economic considerations, and integration of nuanced chemical knowledge. While computer-aided synthesis planning (CASP) tools can generate synthetic routes, evaluating their overall feasibility and quality continues to rely heavily on human expertise, which is often lacking consistency and reproducibility. To address this, we have developed a data-driven scoring model augmented with human expert knowledge. Experts select key synthesis aspects to score the multi-step routes and apply them to the modeling. The model produces target-specific scores for synthetic routes, achieving a top-1 ranking accuracy of 60 \% when benchmarked against experimental data and 90 \% correspondence to the expert synthesis evaluation. Moreover, by incorporating human experts' assessment of routes based on the final route scores, we refined the score into an expert-augmented assessment standard that categorizes routes as Good, Plausible, or Bad universally and interpretably. We demonstrate that this criterion and the resulting route rankings align with expert judgment and synthesis feasibility, that can be obtained from the published reaction data.