Project Ariadne: Prompt-Conditioned Route Generation for Synthesis Planning

2026-06-23 · Source: Takara TLDR - Daily AI Papers · Field: Science & Research — Artificial Intelligence & Machine Learning, Physical Sciences & Chemistry, Research Methodology & Innovation · Depth: Expert, medium

Summary

Project Ariadne is a novel decoder-only route generator designed for retrosynthetic planning, which aims to connect target molecules to commercially available starting materials through multi-step routes. Unlike classical planners or existing direct-generation methods that require separate models for different specifications, Ariadne represents the target, optional constraints, and the entire route within a single prompt-completion sequence. A 24-layer checkpoint of Ariadne demonstrated significant performance improvements on the RetroCast/PaRoutes mkt-cnv-160 benchmark family. Adding prompt fields for route-depth and required-starting-material constraints raised Solv-0 by 13.7 points and 31.2 points, respectively. Ariadne also surpassed DESP, a bidirectional search planner, on required-leaf Top-10 and Solv-0, completing tasks in 24 GPU-minutes versus DESP's 6.8 GPU-hours. While comparable to DMS Explorer XL in standard reconstruction at half the inference time and showing gains on route-holdout reconstruction, AiZynthFinder MCTS still leads on some Solv-0 comparisons. The codebase and training scripts are publicly released.

Key takeaway

For AI Scientists and Research Scientists evaluating retrosynthesis planning tools, Project Ariadne offers a compelling alternative to classical search-based methods. You should consider integrating prompt-conditioned sequence generation models to efficiently handle specific route constraints like depth or required starting materials. This approach significantly reduces computational time, as demonstrated by Ariadne's 24 GPU-minute performance versus 6.8 GPU-hours for DESP, accelerating your synthesis planning workflows.

Key insights

Ariadne uses prompt-conditioned sequence generation to unify retrosynthesis planning, significantly improving constrained route generation efficiency.

Principles

• Retrosynthesis can be reframed as sequence generation.
• Prompt conditioning enhances model adaptability to constraints.
• Direct route generation can outperform search-based methods.

Method

Ariadne is a decoder-only model that generates a complete retrosynthetic route by taking the target molecule and optional constraints as a single prompt-completion sequence.

In practice

• Use prompt conditioning for specific retrosynthesis constraints.
• Explore direct route generation for faster planning.
• Integrate prompt-conditioned models for complex chemical synthesis.

Topics

• Retrosynthesis Planning
• Prompt Conditioning
• Sequence Generation
• Chemical Synthesis
• Decoder-only Models
• Computational Chemistry

Code references

• AMAP-ML/MobilityBench

Best for: AI Scientist, Research Scientist

Related on AIssential

• Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching — Atom ordering in neural representations can encode chemical knowledge, improving retrosynthesis model efficiency and accuracy.
• ARIADNE: Agnostic Routing for Inference-time Adapter DyNamic sElection — ARIADNE enables training-free, adapter-agnostic selection by matching input embeddings to adapter-specific centroids in latent space.
• AlphaJet: Automated Conceptual Aircraft Synthesis via Disentangled Generative Priors and Topology-Preserving Evolutionary Search — AlphaJet automates conceptual aircraft design using generative AI and evolutionary search for real-time 3D synthesis.
• De novo molecular generation with optical property preconditioning at the token level — Token-conditioned autoregressive models can generate OLED molecules with targeted optical properties, but control varies by chemical motif.
• RxnNano:Training Compact LLMs for Chemical Reaction and Retrosynthesis Prediction via Hierarchical Curriculum Learning — Deep chemical understanding and structured training enable compact LLMs to outperform larger models in reaction prediction.
• HiRes: Inspectable Precedent Memory for Reaction Condition Recommendation — HiRes combines retrieval-augmented learning with hierarchical reaction representations to provide accurate, inspectable reaction condition recommendations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.