SynLaD generates synthesizable molecules from 3D pharmacophore profiles

Category: Paper
Status: developing
Sources: arXiv 2607.01105
Summary: SynLaD (Cretu et al., arXiv 2607.01105, dated 2026-07-01, cs.LG) is a latent-diffusion framework for de novo molecule design that jointly targets pharmacophore match and synthesizability. An encoder maps molecules to a shared latent space with two decoder heads, one reconstructing 3D structure and one generating a reaction-based synthetic route, and a diffusion transformer generates novel molecules conditioned on 3D pharmacophore profiles. The authors report outperforming baselines on synthesizable and diverse hit generation.
Why it matters: Coupling shape-conditioned generation with an explicit synthesis route addresses a recurring gap in generative drug design, where proposed molecules are hard to make.
Follow-up: Watch for released code or weights and independent reproduction of the synthesizability and hit-diversity claims.