Researchers have developed a new AI framework called PGFS++, which can improve the properties of molecules in early-stage drug discovery. The framework is based on reinforcement learning and takes into account the synthesis process to produce molecules that are not only effective but also feasible to manufacture. Unlike previous methods, PGFS++ does not rely on indirect reactant selection, making it more efficient. However, the researchers also identified a potential issue with the method, where it can focus too much on improving the reward and sacrifice output diversity. To address this, they introduced a new framework that balances both goals. --- Why it matters: This matters to engineers working in AI because PGFS++ demonstrates how reinforcement learning can be applied to real-world problems like drug discovery, where synthesis constraints play a crucial role. The ability to balance improvement of molecular properties with output diversity is essential for practical applications. Source: https://arxiv.org/abs/2608.19121