Organic chemistry reaction's product can be predicted by a program developed by researchers in IBM. Modelled on the latest language translation systems – like Google’s artificial neural network – the AI picked the right product 80% of the time despite not having been taught any organic chemistry rules.
Teodoro Laino, researcher involved in study at IBM in Zurich Switzerland said, “what program do is trying to imitate a top pro chemist in more or less the entire domain of organic chemistry”. His ambitious goal is shared by other chemists who have been attempting to create a functioning AI chemist since the 1970s, when organic chemist E J Corey kick-started the field by creating a chemical knowledge database.
Laino explained the approach that his team took to create a tool, it can be time-consuming to made tool based on chemistry knowledge. Bartosz Grzybowski’s team took 10 years to encode their Chematica retro-synthesis program with 20,000 chemical rules, he added. There’s a way to learn organic chemistry that’s not memorising chemical rules, by just trying to find out the underlying patterns in reactions and trying to rationalise them, knowledge-based AI has difficulty tackling reactions that lie outside of its rule set.
Philippe Schwaller of the IBM team explain, instead of teaching their program rules, team gave more than 50,000 patented reactions to train on, from reactant plus the reagent, tool tries to guess the most likely product. The same training set showing again and again, tool slowly learns how to construct a valid product. Chemical structures are first converted into a string of letters and numbers. The program then treats the reaction like a translation problem, using the robust algorithms originally developed for language processing.
Program managed to give 80.3% right product of the time, when it was presented with a new set of patented reactions it hadn't encountered before, just after 24 hours of learning. The IBM team says this means its AI outperform a comparable prediction program, created at the Massachusetts Institute of Technology (MIT), US by a margin of 6.3%.
Connor Coley, graduate student and part of the MIT team, said that the IBM team showed a marginal improvement in accuracy and showed that this framework is applicable to this problem. However these kind of models that don't give you an understanding of what actually happens to the chemistry may have a challenge in terms of convincing the chemistry community to accept these “black box” type models’, adds Klavs Jensen, who recently created an AI chemist that combines rule-free learning with some chemical expertise.