REVIEW 11 cited by
Language models can generate molecules, materials, and protein binding sites directly in three dimensions as XYZ, CIF, and PDB files
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Language models are powerful tools for molecular design. Currently, the dominant paradigm is to parse molecular graphs into linear string representations that can easily be trained on. This approach has been very successful, however, it is limited to chemical structures that can be completely represented by a graph -- like organic molecules -- while materials and biomolecular structures like protein binding sites require a more complete representation that includes the relative positioning of their atoms in space. In this work, we show how language models, without any architecture modifications, trained using next-token prediction -- can generate novel and valid structures in three dimensions from various substantially different distributions of chemical structures. In particular, we demonstrate that language models trained directly on sequences derived directly from chemical file formats like XYZ files, Crystallographic Information files (CIFs), or Protein Data Bank files (PDBs) can directly generate molecules, crystals, and protein binding sites in three dimensions. Furthermore, despite being trained on chemical file sequences -- language models still achieve performance comparable to state-of-the-art models that use graph and graph-derived string representations, as well as other domain-specific 3D generative models. In doing so, we demonstrate that it is not necessary to use simplified molecular representations to train chemical language models -- that they are powerful generative models capable of directly exploring chemical space in three dimensions for very different structures.
Forward citations
Cited by 11 Pith papers
-
How Well Can Frontier Large Language Models Generate Structures? High Quality Prediction of Molecular Geometries with Help from Fine-Tuning
A fine-tuned LLM trained on Z-matrix representations generates accurate 3D molecular geometries and conformer ensembles, matching or beating the specialized DMCG model on GEOM benchmarks.
-
Reaction-Network-Level Discovery of Ammonia Synthesis Catalysts via Ten-Million-Scale Generative Exploration
Ten-million-scale generative Transformers with ML potentials map compatibility across N*, NH*, NNH*, and HNNH* to discover 279 ammonia synthesis catalyst candidates, recovering Fe/Ru motifs and identifying new familie...
-
Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints
General-purpose LLMs can satisfy simple local 3D constraints such as anchor fragments and pharmacophores, but their unoptimized poses dock much worse than those from specialized diffusion models.
-
Atomic Design Transformer: Scaffold-Conditioned 3D Molecule Generation via xTB-Reward Reinforcement Learning
A plain causal transformer that tokenizes atom positions in local frames generates 3D molecules directly; RL against an xTB relaxation reward lifts topology-preserving valid yield from ~50% to ~95%.
-
AGAPI-Agents: An Open-Access Agentic AI Platform for Accelerated Materials Design on AtomGPT.org
An open-source agentic materials-design platform shows tool access can help or hurt accuracy depending on the property, but its headline memorization-resistant test results are not presented.
-
InertialAR: Autoregressive 3D Molecule Generation with Inertial Frames
An autoregressive transformer with inertial-frame tokenization and geometric rotary positional encoding reports state-of-the-art validity and stability on QM9, GEOM-Drugs, and B3LYP, plus strong functional-group-condi...
-
CrystalICL: Enabling In-Context Learning for Crystal Generation
CrystalICL is a few-shot crystal generation model combining space-group tokenization with multi-task instruction tuning.
-
Kinetic Langevin Diffusion for Crystalline Materials Generation
KLDM runs the diffusion process for crystal coordinates in Euclidean velocity space via left-trivialized kinetic Langevin dynamics on a torus, and reports competitive or state-of-the-art performance on CSP and DNG benchmarks.
-
3DMolFormer: A Dual-channel Framework for Structure-based Drug Discovery
A dual-channel transformer that reads and writes 3D coordinates as continuous numbers alongside chemical tokens achieves state-of-the-art docking and pocket-aware molecule generation.
-
TopoMAS: Large Language Model Driven Topological Materials Multiagent System
TopoMAS is a multi-agent LLM framework that automates retrieval, generation, and first-principles validation for topological materials, reporting 94.55% accuracy with a lightweight Qwen2.5-72B model.
-
A Survey of AI for Materials Science: Foundation Models, LLM Agents, Datasets, and Tools
This survey organizes foundation models, LLM agents, datasets, and tools in materials science into six task areas.
Discussion (0). Continue with ORCID to comment.