Neural networks taught to propose new types of chemical reactions

Figure: Modeling workflow for generation of new reactions consists of five main steps: (1) training sequence-to-sequence autoencoder on the USPTO database of chemical reactions; (2) building of Generative Topographic Map (GTM) using the autoencoder latent variables and preparation of GTM class landscape; (3) selecting on GTM a zone populated to Suzuki coupling reactions and identification of related autoencoder latent vectors; (4) sampling from the autoencoder latent space and generation of new reactions; and, (5) post-processing step. On the Generative Topographic Map, larger transparency levels correspond to lower density. The color code renders the (binary: Suzuki vs Other) reaction class distribution. Thus, zones in dark blue are exclusively populated by Suzuki reactions, zones in dark red are exclusively populated by other types of reactions; while intermediate colors correspond to reaction space areas hosting both categories, in various ratios. The red circle indicates the zone from which virtual Suzuki reactions were sampled.

An international collaboration has resulted in a paper in Scientific Reports.

Associate Professor Timur Madzhidov, one of the co-authors of the publication, explains, “First, we fed existing chemical reactions to deep neural networks. After that, the networks started generating new types of reactions, which we adjusted specifically to resemble the well-known Suzuki reactions. The AI suggested reactions of various feasibility – from rather reasonable, to trivial, to very brave and even outright improbable. We created a special complicated filter to sort the reactions based on their novelty. Interestingly, the system issued several types of reactions resembling the Suzuki reactions. They were not present in our selection which we fed to the system (consisting of pre-2016 data), but they can be found in more contemporary literature. This proves that the chosen approach can help find new reactions.”

Kazan Federal University was joined by the University of Strasbourg and the University of Hokkaido in this work.

“The Suzuki reaction, commonly used to obtain polyphenols, styrenes, and substituted biphenyls, was chosen for a reason. Akira Suzuki made his discovery at the University of Hokkaido and was eventually awarded the Nobel Prize for it,” adds Madzhidov.

The new software tool will be useful in synthesizing new compounds.

Discovery of novel chemical reactions by deep generative recurrent neural network

https://www.nature.com/articles/s41598-021-81889-y

Source text: Larisa Busil

Translation: Yury Nurmeev