Synthesizability via reward engineering: expanding generative molecular design into synthetic space
Abstract
multi-parameter validation demonstrated clear advantages over naïve, synthesis-unaware baselines: SynthSense generated 6.2-fold more synthetically feasible hits than control trained without SynthSense, achieved a 727-fold enrichment in hits synthesizable with a predefined synthetic route, and populated 4.1-fold more virtual parallel synthesis plates. These results demonstrate that by reframing synthesizability from a mere constraint into an active design objective, generative AI can better support the realities of modern medicinal chemistry by enabling personalized synthetic design, accelerating SAR exploration and aligning more naturally with automated parallel synthesis workflows.
Fields of Study & Subjects
Chemical Synthesis and Analysis
Machine Learning in Materials Science
Computational Drug Discovery Methods
Publication details
- Journal / source
- Chemical Science
- Publisher
- Royal Society of Chemistry (RSC)
- Volume
- 17
- Issue
- 20
- Pages
- 10015–10028
- Publication date
- 1 January 2026
- Language
- EN