Search papers, labs, and topics across Lattice.
SynLaD is a novel latent diffusion framework designed to generate small molecules by integrating ligand-based drug design with synthetic accessibility. By leveraging a dual-decoder architecture that reconstructs both 3D molecular structures and synthetic routes, SynLaD effectively addresses the trade-off between generating high-scoring and synthesizable compounds. The model outperforms existing baselines in generating diverse and synthesizable bioactive ligands, showcasing its potential to streamline drug discovery processes.
A single model can simultaneously generate shape-aligned molecules and feasible synthesis plans, overcoming the traditional trade-offs in drug design.
We present SynLaD, a latent diffusion framework for small-molecule generation that unifies ligand-based drug design objectives (what to make) with synthetic accessibility (how to make it). Current models typically optimize one objective at the expense of the other, creating a bottleneck for discovering high-scoring and synthesizable molecules. SynLaD combines reaction-constrained generation with pharmacophore-conditioned 3D design by learning a latent space that decodes to both 3D structures and synthesis pathways. An encoder maps molecules to a latent representation used by two decoder heads: (i) a geometric head that reconstructs atom types and coordinates and (ii) an autoregressive synthesis head that outputs synthetic routes in a serialized, reaction-based notation. A diffusion transformer generates novel latents in the learned space, conditioned on pharmacophore profiles. Across analogue generation tasks for bioactive ligands, SynLaD outperforms existing baselines in synthesizable and diverse hit generation, demonstrating that a single model can produce shape-aligned molecules with feasible synthesis plans.
