Just out: Generative pretraining for drug molecule design with bidirectional structure-property optimization
Communications Chemistry, Published online: 08 June 2026; doi:10.1038/s42004-026-02093-5Designing drug-like molecules that satisfy both property requirements and structural constraints is a persistent challenge. Here, the authors introduce BiSP-GP, a bidirectional structure-property generative pretraining framework that unifies molecular generation and property prediction through shared autoregressive sequence modeling, enabling joint learning and bidirectional optimization of structure-property relationships, thereby enhancing controllable molecular generation while supporting molecular property prediction and candidate drug optimization tasks.
