Project Ariadne: Prompt-Conditioned Route Generation for Synthesis Planning

Researchers introduce Ariadne, a decoder-only model that reframes retrosynthetic planning as prompt-conditioned sequence generation, allowing target molecules, constraints, and routes to be represented in a single sequence. This approach eliminates the need for separate models tailored to specific planning specifications.

• On the RetroCast/PaRoutes mkt-cnv-160 benchmark, adding prompt fields for route depth and required starting materials increased Solv-0 scores by 13.7 and 31.2 points respectively.
• Ariadne outperforms the DESP bidirectional search planner on required-leaf Top-10 and Solv-0 metrics while requiring only 24 GPU-minutes compared to 6.8 GPU-hours.
• The model achieves reconstruction performance comparable to DMS Explorer XL at approximately half the reported inference time.
• While Ariadne shows gains in route-holdout reconstruction, AiZynthFinder MCTS remains stronger on several Solv-0 comparisons across target-only benchmarks.

The authors release the codebase and training scripts to support further development, noting that the absence of Tier-1--3 route checkers remains the primary bottleneck for experimental utility.