Search papers, labs, and topics across Lattice.
This paper introduces Ariadne, a novel decoder-only model for prompt-conditioned retrosynthetic route generation that integrates target molecules, constraints, and routes into a single prompt-completion sequence. By leveraging this unified approach, Ariadne significantly outperforms existing methods on the RetroCast/PaRoutes benchmark, achieving notable improvements in solvability scores while drastically reducing computational time compared to traditional planners. The results indicate that Ariadne effectively extends sequence generation techniques to structural route generation, although further advancements are needed to address the limitations of current route-checking mechanisms.
Ariadne achieves a 31.2-point increase in solvability for constrained retrosynthesis while cutting computation time from hours to minutes.
Retrosynthetic planning seeks to connect a target molecule to commercially available starting materials through a multistep route. Classical planners construct such routes by iteratively applying single-step reaction models within a search procedure; constrained variants often require specialized algorithms or architectural changes. Direct route generation reframes retrosynthesis as sequence generation, but existing direct-generation methods still train separate models for different planning specifications. We introduce Ariadne, a decoder-only route generator that represents the target, optional constraints, and route in one prompt-completion sequence. On the RetroCast/PaRoutes mkt-cnv-160 benchmark family, one 24-layer checkpoint follows route-depth and required-starting-material prompts: adding the corresponding prompt fields raises Solv-0 by 13.7 points for depth constraints and 31.2 points for required-leaf constraints. Ariadne also improves over DESP, a bidirectional search planner, on required-leaf Top-10 and Solv-0 in 24 GPU-minutes versus 6.8 GPU-hours. On standard reconstruction, Ariadne is comparable to DMS Explorer XL at about half the reported inference time. Across additional target-only benchmarks, Ariadne's clearest gains are on route-holdout reconstruction, whereas AiZynthFinder MCTS remains stronger on several Solv-0 comparisons. These results extend sequence generation from specialist retrosynthesis models to prompt-conditioned structural route generation. We release the codebase and training scripts to support further work, but do not introduce Tier-1--3 route checkers; those remain the main bottleneck before models of this kind can become useful to experimental chemists.
