Machine Learning for Organic Synthesis

This lecture discusses the progress towards leveraging machine learning for organic synthesis, focusing on predicting reaction yields using databases, high throughput experimentations, and the importance of dataset predictability. It also covers the NiCOlit dataset as a proof of concept, steps towards yield prediction, and predictive performances on random splitting. The comparison of NiCOlit and HTE datasets, including chemical space size, diversity, and scope/optimization structure, is analyzed. The lecture concludes with insights on the behavior of out-of-sample predictions, visualization of publications distribution in reaction space, and perspectives for predictive chemistry.