This paper proposes a flexible zero-shot molecular manipulation method by exploring the shared latent space of 3D molecules. We introduce a variational autoencoder (VAE) for 3D molecules, called MolFLAE, which learns a fixed-dimensional E(3) equilateral latent space, independent of the number of atoms. MolFLAE uses an E(3) equilateral neural network to encode 3D molecules into a fixed number of latent nodes, each distinguished by a learned embedding. The latent space is normalized, and molecular structures are reconstructed via a Bayesian flow network (BFN) conditioned on the latent output of the encoder. MolFLAE achieves competitive performance on standard unconditional 3D molecule generation benchmarks. Furthermore, MolFLAE's latent space enables zero-shot molecular manipulation, including atom count editing, structural reconstruction, and coordinated latent interpolation for both structure and properties. Finally, we demonstrate our approach on a drug optimization task for the human glucocorticoid receptor, generating molecules with enhanced hydrophilicity while preserving key interactions under computational evaluation.
