Bridging the Gap between Learning and Inference for Diffusion-Based Molecule Generation

Drug discovery entails a comprehensive understanding of the molecular underpinnings of disease pathophysiology, followed by the identification and synthesis of chemical entities or biopharmaceuticals capable of selectively modulating the pertinent biological pathways (Sneader, 2005). Among the numerous traditional methods, screening from natural products and serendipitous discoveries are the most renowned. The discovery of penicillin and artemisinin (White, 1997), two antibiotics, relied on the former method, while the drug repurposing of sildenafil (Eardley et al., 2002) for the treatment of erectile dysfunction owed to the latter approach. Subsequently, new biologybased and computer-assisted methods have achieved encouraging results (Mandal et al., 2009; Rognan, 2007; Batool et al., 2019). For instance, rational drug design lowers the overall cost by targeting known protein pockets, and highthroughput screening (Mayr and Bojanic, 2009) enables faster identification of molecules with potential drug activity.