Positivity is All You Need (PAYN): A PU Learning Framework for Yield Prediction in Organic Chemistry

Accurately predicting chemical reaction yields in silico is a long-standing goal in organic chemistry that, if achieved, would revolutionize synthesis design, op-timization, and discovery. The vast reaction data within scientific literature rep-resents a rich resource for training predictive machine learning models, howev-er, this resource is fundamentally compromised by a pervasive selection and re-porting bias, resulting in imbalanced datasets. Here, we introduce "Positivity is All You Need" (PAYN), a machine learning framework that addresses this data-scarcity problem by learning directly from biased, positive-only data. PAYN leverages Positive-Unlabeled (PU) learning, treating reported high-yielding re-actions as the ‘positive’ class and the vast, unexplored chemical space as the ‘unlabeled’ class. To validate our approach, we simulated literature bias on fully labeled High-Throughput Experimentation (HTE) datasets. We demonstrated that PAYN can significantly improve the performance of models trained on lit-erature data, by balancing the data with augmented negative datapoints. This work establishes a robust framework for leveraging biased historical data, pav-ing a path toward more scalable and accessible data-driven strategies for accel-erating synthesis design, optimization and chemical discovery.