# Coverage Before Control: Route-Instruction Grounding and Steering for Controllable Retrosynthesis

arXiv: https://arxiv.org/abs/2609.39955 · HTML: https://mlsanity.com/paper/2609.39955/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 46 references that are in the map.
- First posted on arXiv 30 Sep 2026.

## Builds on (most cited first, 23 of 23)

- [Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models](https://mlsanity.com/paper/2506.05176/) (2025)
- [Learning to Plan Chemical Syntheses](https://mlsanity.com/paper/1708.04202/) (2017)
- [Translation between Molecules and Natural Language](https://mlsanity.com/paper/2204.11817/) (2022)
- [Multi-modal Molecule Structure-text Model for Text-based Retrieval and Editing](https://mlsanity.com/paper/2212.10789/) (2022)
- [Simple Guidance Mechanisms for Discrete Diffusion Models](https://mlsanity.com/paper/2412.10193/) (2024)
- [Mol-Instructions: A Large-Scale Biomolecular Instruction Dataset for Large Language Models](https://mlsanity.com/paper/2306.08018/) (2023)
- [Retro*: Learning Retrosynthetic Planning with Neural Guided A* Search](https://mlsanity.com/paper/2006.15820/) (2020)
- [RetroBridge: Modeling Retrosynthesis with Markov Bridges](https://mlsanity.com/paper/2308.16212/) (2023)
- [Double-Ended Synthesis Planning with Goal-Constrained Bidirectional Search](https://mlsanity.com/paper/2407.06334/) (2024)
- [Learning to Make Generalizable and Diverse Predictions for Retrosynthesis](https://mlsanity.com/paper/1910.09688/) (2019)
- [Chemical reasoning in LLMs unlocks strategy-aware synthesis planning and reaction mechanism elucidation](https://mlsanity.com/paper/2503.08537/) (2025)
- [AOT*: Efficient Synthesis Planning via LLM-Empowered AND-OR Tree Search](https://mlsanity.com/paper/2509.20988/) (2025)
- [Synthelite: Chemist-aligned and feasibility-aware synthesis planning with LLMs](https://mlsanity.com/paper/2512.16424/) (2025)
- [Equivariant Denoisers Cannot Copy Graphs: Align Your Graph Diffusion Models](https://mlsanity.com/paper/2405.17656/) (2024)
- [RetroDiff: Retrosynthesis as Multi-stage Distribution Interpolation](https://mlsanity.com/paper/2311.14077/) (2023)
- [Preference Optimization for Molecule Synthesis with Conditional Residual Energy-based Models](https://mlsanity.com/paper/2406.02066/) (2024)
- [RETRO SYNFLOW: Discrete Flow Matching for Accurate and Diverse Single-Step Retrosynthesis](https://mlsanity.com/paper/2506.04439/) (2025)
- [Retrosynthesis prediction enhanced by in-silico reaction data augmentation](https://mlsanity.com/paper/2402.00086/) (2024)
- [When Single Answer Is Not Enough: Rethinking Single-Step Retrosynthesis Benchmarks for LLMs](https://mlsanity.com/paper/2602.03554/) (2026)
- [T-Rex: Text-assisted Retrosynthesis Prediction](https://mlsanity.com/paper/2401.14637/) (2024)
- [Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching](https://mlsanity.com/paper/2602.13136/) (2026)
- [From Feasible to Practical: Pareto-Optimal Synthesis Planning](https://mlsanity.com/paper/2605.07521/) (2026)
- [Project Ariadne: Prompt-Conditioned Route Generation for Synthesis Planning](https://mlsanity.com/paper/2606.24184/) (2026)

## Abstract (arXiv)

Single-step retrosynthesis models are commonly evaluated by their ability to recover recorded reactions. In practice, chemists may need to choose among several precursor sets for the same product, for example to preserve a particular motif. Recovering a recorded answer alone does not establish this ability to follow a preference. Satisfying such requests requires both coverage of relevant alternatives and control over which alternatives are favored. We introduce Route-Instruction Grounding and Steering (RIGS), a two-stage framework for instruction-conditioned retrosynthesis. Stage A trains a language projector, teaching it which alternatives an instruction favors or discourages. Stage B uses the projector learned in Stage A to steer a frozen generative model through lightweight residual adapters. We construct nested one-to-many training supports by pairing each product with increasing numbers of candidate precursor sets. Extensive experiments demonstrate that broader support helps the model generate a wider range of alternatives, and RIGS can learn to guide generation according to instructions. The relationship between coverage and control is consistent across model scales but non-monotone.