AI Learns Chemistry to Predict How to Make Medicines

Researchers have invented a machine learning algorithm that forecasts the outcome of chemical reactions with much greater accuracy than proficient chemists and proposes ways to make complex particles, removing a weighty hurdle in drug innovation. University of Cambridge researchers have revealed that an algorithm can forecast the outcomes of complex chemical reactions with above 90% accuracy, overtaking trained chemists. The algorithm also displays chemists how to make target mixtures, providing the chemical "map" to the required destination. The outcomes are stated in two studies in the journals ACS Central Science and Chemical Communications. A central task in drug discovery and materials science is discovering ways to build complex organic molecules by chemically assembling together elementary building blocks. The problem arises when those building blocks often respond in surprising ways.