Fixed dimensional model generates 3D molecules of varying sizes
Fixed-Dimensional Latent Flow for Generating Variable-Size 3D Molecules
Machine Learning
Summary
Generating 3D molecules is tricky because their size, shape, and properties are all connected. The authors created a new method that uses a fixed-size code to represent molecules, no matter their size. This code then helps generate molecules with varying sizes and chemical details accurately. Their approach avoids complex math for symmetry and improves how well the generated molecules match real ones, including their chemical properties. It also speeds up the creation process and can focus on making molecules with desired features.
What this means in practice
- •For drug discovery teams: Generate diverse 3D molecular candidates of variable sizes with reliable chemical structures faster for screening and property optimization.
- •For material design engineers: Produce novel molecules with targeted electronic properties by conditioning molecular generation on specific property values using the same fixed-size code.
Authors
Weichi Yao, Cameron Gruich, Bryan R. Goldsmith, Yixin Wang
Abstract
In molecular discovery, molecule size is coupled to composition, structure, and other target properties. Yet most 3D generators require molecule size to be specified before generation. Here, we introduce Equivariant-Free Transformer-Autoencoded Latent Flow Matching, a two-stage generative framework that relies entirely on a single fixed-dimensional molecule-level latent representation to generate variable-size molecules. The second-stage flow matching model samples this latent vector, and an autoregressive Transformer decoder then determines molecule size while generating atom types, coordinates, and chemically informative states. Canonical atom ordering and rigid-pose alignment enable standard Transformers without equivariant layers, while joint decoding of molecular geometry and an enriched chemical state enables reliable, deterministic, chemistry-guided graph recovery without requiring a learned dense pairwise bond decoder. The same fixed-dimensional latent supports unconditional and property-conditioned flow matching, while optional property supervision adds an internal ranking readout, with no separate predictor or reference calculations. On PCQM4Mv2, EF-TALFM achieves the highest fraction of molecules that are unique, training-set novel, pass sanitization and PoseBusters sanity checks, 89.4\%, compared with 75.6\% for UAE-3D and 69.8\% for FlowMol. EF-TALFM also achieves higher measured computational throughput for training and sampling. Across ten target HOMO--LUMO gaps, internal ranking doubles the density functional theory (DFT)-verified hit rate within $0.1\,\mathrm{eV}$, while preserving 97\% novelty among unique verified hits. These results demonstrate that fixed-dimensional molecule-level generation followed by symmetry-resolved autoregressive realization provides a practical architecture for open-ended and property-directed 3D molecular design.