mlsanityData to 14 Oct 2026

AtomComposer: Discovering Chemical Space from First Principles with Reinforcement Learning

arXiv 2605.28287 ↗First posted 27 May 2026cs.LG cond-mat.mtrl-scicited by 1

On the map: Deep reinforcement learning methods → RL for energy, networks and operations

Abstract (arXiv)

Discovering novel stable molecules without training data remains a grand scientific challenge. Current molecular generative models are trained on large, pre-curated datasets, which introduce biases and limit exploration of novel chemistry. In contrast, we propose a new paradigm: autonomous, generalized agents capable of mapping vast, unknown chemical spaces without any pretraining. For the first time, we present AtomComposer, a self-guided agent that autonomously constructs valid 3D isomers under stoichiometric constraints and is trained exclusively online using reinforcement learning. Unlike existing approaches that generally overfit to a specific chemical formula, we establish a multi-composition training scheme that enables a broad generalization across diverse chemistry, guided by energy- and validity-based rewards. Our agent can discover up to an order of magnitude more valid isomers on unseen test formulas than existing single-composition reinforcement-learning baselines trained with per-step energy rewards. These results fulfill the promise of online reinforcement learning as a powerful paradigm for scalable, from-scratch exploration of chemical configuration space.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in RL for energy, networks and operations, one of 15 directions in Deep reinforcement learning methods. Builds on 38 of its 82 references that are in the map. First posted on arXiv 27 May 2026.

Builds on

38 of 82 references.

References · 38

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2018Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actorcited by 4,885
  3. 2015High-Dimensional Continuous Control Using Generalized Advantage Estimationcited by 2,026
  4. 2023Training Diffusion Models with Reinforcement Learningcited by 982
  5. 2016Automatic chemical design using a data-driven continuous representation of moleculescited by 765
  6. 2017Deep Reinforcement Learning that Matterscited by 754
  7. 2022Equivariant Diffusion for Molecule Generation in 3Dcited by 677
  8. 2023Safe RLHF: Safe Reinforcement Learning from Human Feedbackcited by 656
  9. 2017SchNet: A continuous-filter convolutional neural network for modeling quantum interactionscited by 571
  10. 2021Equivariant message passing for the prediction of tensorial properties and molecular spectracited by 441
  11. 2018Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generationcited by 364
  12. 2018GuacaMol: Benchmarking Models for De Novo Molecular Designcited by 297
26 more
  1. 2018Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Modelscited by 275
  2. 2017Molecular De Novo Design through Deep Reinforcement Learningcited by 267
  3. 2021A Practical Guide to Multi-Objective Reinforcement Learning and Planningcited by 256
  4. 2018Optimization of Molecules via Deep Reinforcement Learningcited by 200
  5. 2019Symmetry-adapted generation of 3d point sets for the targeted discovery of moleculescited by 172
  6. 2017Deep Reinforcement Learning for De-Novo Drug Designcited by 161
  7. 2021MARS: Markov Molecular Sampling for Multi-objective Drug Discoverycited by 125
  8. 2017Reproducibility of Benchmarked Deep Reinforcement Learning Tasks for Continuous Controlcited by 93
  9. 2021Inverse design of 3d molecular structures with conditional generative neural networkscited by 93
  10. 2020An empirical investigation of the challenges of real-world reinforcement learningcited by 77
  11. 2020Symmetry-Aware Actor-Critic for 3D Molecular Designcited by 55
  12. 2020Reinforcement Learning for Molecular Design Guided by Quantum Mechanicscited by 51
  13. 2023Navigating the Design Space of Equivariant Diffusion-Based Generative Models for De Novo 3D Molecule Generationcited by 50
  14. 2024SemlaFlow -- Efficient 3D Molecular Generation with Latent Attention and Equivariant Flow Matchingcited by 41
  15. 2022A Systematic Survey of Chemical Pre-trained Modelscited by 34
  16. 2019A Hitchhiker's Guide to Statistical Comparisons of Reinforcement Learning Algorithmscited by 27
  17. 2019Generative Models for Automatic Chemical Designcited by 23
  18. 2020A reinforcement learning approach to rare trajectory samplingcited by 22
  19. 2023Symphony: Symmetry-Equivariant Point-Centered Spherical Harmonics for 3D Molecule Generationcited by 17
  20. 2020Machine learning of solvent effects on molecular spectra and reactionscited by 14
  21. 2023Towards equilibrium molecular conformation generation with GFlowNetscited by 14
  22. 2025Equivariant Neural Diffusion for Molecule Generationcited by 13
  23. 2022Scalable Fragment-Based 3D Molecular Design with Reinforcement Learningcited by 11
  24. 2021Generating stable molecules using imitation and reinforcement learningcited by 9
  25. 2019Atomistic structure learningcited by 5
  26. 2020Atomistic Structure Learning Algorithm with surrogate energy model relaxationcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Information-Dense Synthesis for Molecular Discoverycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Guiding Generative Models to Uncover Diverse and Novel Crystals via Reinforcement Learning
  2. 2025Guiding Evolutionary Molecular Design: Adding Reinforcement Learning for Mutation Selection
  3. 2024It Takes Two to Tango: Directly Optimizing for Constrained Synthesizability in Generative Molecular Design
  4. 2025Test-Time Training Scaling Laws for Chemical Exploration in Drug Design
  5. 2026MolEvolve: LLM-Guided Evolutionary Search for Interpretable Molecular Optimization
  6. 2026Agentic reinforcement learning empowers next-generation chemical language models for molecular design and synthesis
  7. 2025ATOM: A Pretrained Neural Operator for Multitask Molecular Dynamics
  8. 2024Generative artificial intelligence for computational chemistry: a roadmap to predicting emergent phenomena