AI can accurately predict reaction yield and is expected to be used in new drug development March 09, 2018 Source: WuXi PharmaTech Artificial intelligence (AI) is a topic of explosion in recent years. In the pharmaceutical industry, the debate about whether AI can effectively design new drugs has never stopped. Recently, a study published in the top academic journal Science has shown us a positive side: chemists from Princeton University and Merck Research Laboratories have developed a Artificial intelligence can accurately predict the yield of chemical reactions, which is expected to be widely used in the development of new drugs. In essence, the synthesis and manufacture of new drugs is a chemical reaction. In order to get the molecules they want, chemists must add the right ingredients in the right proportions and provide the right reaction conditions. This sounds simple, but it is not easy for the machine to predict and design high-yield chemical reactions. A typical chemical reaction requires the participation of a large number of different chemical molecules. Each additional numerator is added to add a dimension to the calculation. Even for artificial intelligence that has shown great performance in recent years, chemists are not sure whether it is qualified for the job. In order for the machine to learn to design a high-yield chemical reaction, we need to help it build a multi-dimensional model. But historically, this has always been a bottleneck. The reason behind this is that the model is too complex, and the data we can use to train this model is not enough. Fortunately, with Merck's massive investment in robotics, we are now able to complete thousands of chemical reactions in just a few days. This produces a large amount of data that can be used to train the AI. â–² Dr. Spencer Dreher of the Merck Research Laboratory (Source: Merck's official website) “I am very pleased to see that the data we generate is of high quality and can be used to build effective models,†said Dr. Spencer Dreher of Merck Research Laboratories. After obtaining the data and using the program to quantify each chemical entered into the model, the researchers considered the accuracy of multiple statistical models. Interestingly, the linear regression models commonly used in the industry do not perform well in the execution of this task, and the best performer is a model called “random forestâ€. Such models can randomly extract a small number of samples from the training database to construct a decision tree, and each decision tree can predict the specific chemical reaction yield. These forecasts are evaluated comprehensively to produce an overall predicted yield. The results show that this model can well predict the yield of chemical reactions outside the training data. â–² “Random Forest†model performs best (Source: Science) â–²The model can accurately predict the yield (Source: Science) “With just a few hundred reaction data, we can accurately predict the yield with the model. Chemists can even do this without the help of a robot,†said Dr. Derek Ahneman, the lead author of the study. â–²Professor Abigail Doyle of Princeton University (Source: Princeton University) “The software we developed can be applied to any reaction or any substrate type,†said Professor Abigail Doyle of Princeton University, one of the co-authors of the study. “We hope people can apply this tool to other reactions. ." Reference materials: [1] Chemists harness artificial intelligence to predict the future of chemical reactions [2] Predicting reaction performance in C–N cross-coupling using machine learning Precision Seeding Machine,High Capacity Wheat Planter,Wheatgrass Sowing Machiner,Agricultural Precision Planter Taizhou Yingtian Agricultural Machinery Manufacturing Co., Ltd. , https://www.sakuradaagc.com
AI can accurately predict reaction yield and is expected to be used in new drug development