Generative Drugs Fail the Dose Test

🔥 What happened Maria Franz and N. Benjamin Erichson re-scored 60 shortlisted molecules from three PARP1 generative pipelines using PBPK simulations. Seven never reach target exposure, and 28% of the rest carry predicted safety risks at their required dose. 💡 Why it matters 85% of candidates with an achievable dose get beaten by molecules that static selection had thrown out. Filtering on potency and toxicity alone optimizes for a patient who doesn't exist. ⚡ Our take Generative drug design has been high on benchmarks that quietly ignore dose. This paper is the sober reality check — and a warning shot for any pipeline that doesn't bake PBPK in from day one.