Atomic Policy Optimization is proposed, a fully unsupervised alignment framework that eliminates the need for ground-truth reference structures and suggests that intrinsic physical consistency can serve as a superior guide for alignment compared to noisy, supervised coordinate matching.
Abstract
Predicting the 3D structures of atomic systems is fundamental to advancing material science and drug discovery. While flow-matching models (, FlowDPO) have recently shown promise in this domain, their performance relies heavily on alignment with ground-truth coordinates via supervised preference learning. However, obtaining experimental labels for novel crystal phases or de novo proteins is prohibitively expensive, creating a bottleneck for structural modeling in data-scarce regimes. In this work, we propose (Atomic Policy Optimization), a fully unsupervised alignment framework that eliminates the need for ground-truth reference structures. APO adapts group-relative policy optimization to 3D atomic environments, utilizing a novel dual-reward mechanism: (i) a that reinforces the policy's dominant latent structural modes through eigen-decomposition of sample similarities, and (ii) a that enforces thermodynamic stability. Our framework enables the model to ``self-correct''by identifying physically plausible configurations within sampled groups. Extensive benchmarks on crystal and antibody structure prediction demonstrate that APO consistently outperforms fully supervised baselines, achieving a new state-of-the-art in match rates and structural fidelity. Furthermore, we show that APO effectively straightens probability paths, significantly improving inference efficiency. Our results suggest that intrinsic physical consistency can serve as a superior guide for alignment compared to noisy, supervised coordinate matching.
Crystal Structure Prediction (CSP), the task of determining stable atomic arrangements from chemical composition alone, remains a central challenge in computational materials science with direct implications for accelerating materials discovery. While recent diffusion-based generative models achieve impressive results by conditioning on space-group symmetry information, this paradigm exposes three fundamental challenges: (1) unreliable symmetry inference from composition, (2) reliance on symmetry-only priors without comprehensive structural geometric guidance, and (3) prior-induced distribution shift caused by inaccurate or overly strong constraints. To address these challenges, we propose TGD-CSP, a three-stage generative framework that: (1) learns a cross-modal embedding space to retrieve structurally relevant templates directly from composition, thereby providing reliable symmetry priors; (2) guides diffusion-based generation via score-based conditioning that explicitly incorporates comprehensive geometric information from retrieved templates; and (3) fine-tunes the generative policy via reinforcement learning with an energy-based reward to alleviate prior-induced distribution shift and mitigate biased generation. % Evidence: Concrete results TGD-CSP achieves match rates of 74.56% and 75.25% on Perov-5 and MP-20, respectively, including a 42% relative improvement over state-of-the-art methods on Perov-5. It reduces RMSE to 0.0259 and 0.0258, respectively, while yielding structures with formation energies closely matching those of ground-truth structures. Our results demonstrate that TGD-CSP enables reliable generation from composition alone, significantly improving the practicality of generative models for real materials discovery.
Lu Yang, Tiantian Xu, X. Liu et al.· Proceedings of the 32nd ACM...· 0 citations
GeoNet is a physicochemical-principle-guided framework for modeling dual-range atomic interactions that achieves the smallest model size and the shortest training time, demonstrating both superior predictive performance and computational efficiency.
Packora is presented, a flow-based generative model for molecular CSP that jointly predicts atomic coordinates and the lattice from molecular graphs that outperforms the baselines on both structure generation and ranking benchmarks.
Nayoung Kim, Kiyoung Seong, Sungsoo Ahn· 0 citations
Crystal structure prediction (CSP) is a cornerstone technology for the efficient discovery and rational design of functional materials. Here, we propose a deep-learning-enabled dual-mode CSP framework that simultaneously supports two complementary tasks: predicting stable crystal structures for given elemental compositions and identifying chemically viable elemental substitutions for a predefined crystal topology. The framework employs an improved normalized structural fingerprint descriptor that integrates local coordination topology without global spatial-scale information. A cascaded site-probability model, composed of an autoencoder and sigmoid-based classifiers, is developed to predict the occupancy probabilities of 84 chemical elements at crystallographically distinct sites and to efficiently rank candidate structures accordingly. Remarkably, even when trained solely on topological information, the model autonomously captures elemental chemical similarity and intrinsic periodic trends, yielding chemically meaningful multielement probability distributions. Application of the proposed framework to high-throughput screening of superhard materials in the B–N system successfully identifies a thermodynamically, mechanically, and dynamically stable hexagonal BN phase, together with a metastable monoclinic B2N3 structure. Furthermore, elemental substitution screening on the zinc-blende prototype reveals two metastable compounds, In4Sb4 and Ga4Sb4. The proposed framework provides a low-cost, high-efficiency, data-driven strategy for the rapid discovery of inorganic materials.
This work not only establishes a pioneering paradigm for interpretable ML-driven force field refinement but also provides the first feature engineering solution incorporating chemical, physical, and structural information specifically designed for the machine learning of energetic molecular crystals.
Qi He, Pengju Wang, Xudong He et al.· Molecules· 0 citations
A clear pattern is revealed in LLM spatial capabilities: while they still lag behind state-of-the-art approaches, they are promising and can handle multiple spatial constraints simultaneously, enabling scaling to heterogeneous setups.
Thomas MacDougall, Maksim Kuznetsov, Roman Schutski et al.· arXiv.org· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
