Symphony: Symmetry-Equivariant Point-Centered Spherical Harmonics for 3D Molecule Generation
This work addresses 3D molecular geometry generation for computational chemistry, with incremental improvements over prior autoregressive methods.
The authors tackled 3D molecule generation by introducing Symphony, an E(3)-equivariant autoregressive model that uses spherical harmonic signals, and it outperformed existing autoregressive models on the QM9 dataset, approaching diffusion model performance.
We present Symphony, an $E(3)$-equivariant autoregressive generative model for 3D molecular geometries that iteratively builds a molecule from molecular fragments. Existing autoregressive models such as G-SchNet and G-SphereNet for molecules utilize rotationally invariant features to respect the 3D symmetries of molecules. In contrast, Symphony uses message-passing with higher-degree $E(3)$-equivariant features. This allows a novel representation of probability distributions via spherical harmonic signals to efficiently model the 3D geometry of molecules. We show that Symphony is able to accurately generate small molecules from the QM9 dataset, outperforming existing autoregressive models and approaching the performance of diffusion models.