mlsanityData to 14 Oct 2026

Single-step Retrosynthesis via Reinforcement Learning from Forward Feedback

arXiv 2610.15141 ↗First posted 10 Oct 2026cs.AIcited by 0

On the map: LLM alignment / RLHF / preference optimisation → Reward modeling and RLHF also fits: Preference optimization for molecular design

also in 1 more
Abstract (arXiv)

Single-step retrosynthesis is a one-to-many inverse problem, yet most models are trained and evaluated against a single recorded reactant set. Reference-based training gives no explicit positive signal for unrecorded alternatives, and exact-match evaluation rejects non-matching predictions regardless of chemical plausibility. We propose Retrosynthesis via Reinforcement Learning from Forward Feedback (RetroRLFF), which optimizes reactant generation using both reference agreement and forward consistency with the target product. Starting from a supervised decoder-only Transformer, RetroRLFF generates complete reactant candidates and uses a frozen forward reaction model to predict products. A sequence-level reward combines reference agreement and forward consistency while penalizing invalid outputs. Direct Advantage-Based Policy Optimization updates the policy without human preference annotations or a preference-trained reward model. On USPTO-50K, RetroRLFF outperforms all 14 baselines in candidate-level round-trip accuracy at k = 1, 3, 5, and 10. With approximately 50 million parameters, it achieves 90.1% top-1 round-trip accuracy, exceeding R-SMILES, the strongest competing baseline on this metric, by 8.5 percentage points. Compared with its supervised policy, round-trip accuracy increases from 83.1% to 90.1%, while exact-match accuracy is maintained at 57.9% versus 57.5%. In an anonymized expert assessment of 131 randomly selected test cases, organic synthesis experts prefer RetroRLFF over EditRetro in 67.9% of comparisons. These results support forward feedback as a training signal for improving forward consistency while retaining top-1 reference recovery. Expert assessment provides complementary evidence of synthetic feasibility on the selected cases, although neither forward consistency nor expert preference establishes experimental feasibility.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reward modeling and RLHF, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 25 of its 44 references that are in the map. First posted on arXiv 10 Oct 2026.

Builds on

25 of 44 references.

Closest ancestors · 3

Same direction, newest first

  1. 2023RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback1 Sep
  2. 2022Scaling Laws for Reward Model Overoptimization19 Oct
  3. 2017Deep reinforcement learning from human preferences12 Jun

Other references · 22

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2017Proximal Policy Optimization Algorithmscited by 15,215
  3. 2020Learning to summarize from human feedbackcited by 2,588
  4. 2018Molecular Transformer - A Model for Uncertainty-Calibrated Chemical Reaction Predictioncited by 229
  5. 2020Retrosynthesis Prediction with Conditional Graph Logic Networkcited by 104
  6. 2020Molecule Edit Graph Attention Network: Modeling Chemical Reactions as Sequences of Graph Editscited by 88
  7. 2017Retrosynthetic reaction prediction using neural sequence-to-sequence modelscited by 86
  8. 2020A Graph to Graphs Framework for Retrosynthesis Predictioncited by 86
  9. 2020RetroXpert: Decompose Retrosynthesis Prediction like a Chemistcited by 68
  10. 2021Permutation invariant graph-to-sequence model for template-free retrosynthesis and reaction predictioncited by 68
  11. 2020Learning Graph Models for Retrosynthesis Predictioncited by 65
  12. 2025Training a Scientific Reasoning Model for Chemistrycited by 56
10 more
  1. 2022Root-aligned SMILES: A Tight Representation for Chemical Reaction Predictioncited by 51
  2. 2023RetroBridge: Modeling Retrosynthesis with Markov Bridgescited by 36
  3. 2023Retrosynthesis Prediction with Local Template Retrievalcited by 16
  4. 2024RetroGFN: Diverse and Feasible Retrosynthesis using GFlowNetscited by 12
  5. 2022G2GT: Retrosynthesis Prediction with Graph to Graph Attention Neural Network and Self-Trainingcited by 6
  6. 2023RCsearcher: Reaction Center Identification in Retrosynthesis via Deep Q-Learningcited by 6
  7. 2025RETRO SYNFLOW: Discrete Flow Matching for Accurate and Diverse Single-Step Retrosynthesiscited by 6
  8. 2023RLSynC: Offline-Online Reinforcement Learning for Synthon Completioncited by 4
  9. 2024Retrosynthesis Prediction via Search in (Hyper) Graphcited by 2
  10. 2025Round-trip Reinforcement Learning: Self-Consistent Training for Better Chemical LLMscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025RetroDFM-R: Reasoning-Driven Retrosynthesis Prediction with Large Language Models via Reinforcement Learning
  2. 2026RetroReasoner: A Reasoning LLM for Strategic Retrosynthesis Prediction
  3. 2026An Agentic Retrobiosynthesis Framework with Learned Frontier Selection
  4. 2026Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching
  5. 2025Retro-Expert: Collaborative Reasoning for Interpretable Retrosynthesis
  6. 2023RetroDiff: Retrosynthesis as Multi-stage Distribution Interpolation
  7. 2026Reinforced Reasoning for End-to-End Retrosynthetic Planning
  8. 2025DiffER: Categorical Diffusion for Chemical Retrosynthesis