Obtaining an accurate representation of reaction profiles remains one of the most important challenges in chemistry, delaying the development and optimization of novel solutions to many of the pressing issues facing today?s society, such as the mitigation of air pollutants, the valorization of waste products, the need for advanced energy materials and the establishment of economically and environmentally sustainable chemical processes. Unfortunately, the establishment of precise reaction mechanisms is a huge undertaking for both theoretical chemists and experimentalists, making the mere optimization of reaction conditions an outstanding challenge, thus hindering the design of optimal catalysts for specific applications.
In order to overcome such limitations, REALM - REActive Learning Machines - will apply Machine Learning (ML) methods to explore the chemical space, taking into account the characteristics of the reactants, (expected) products, and reaction conditions such as the presence and nature of the catalyst, and the chemical environment (temperature, solvent, etc). The ML methods selected (Artificial Neural Networks, ANN, and Support Vector Machines, SVM) will be trained using already available experimental data, as well as the results of new computational simulations to be pursuit within the project. These ML applications will also benefit from the feedback provided by experimental verification of their predictions, which will be used to refine their accuracy and also diagnose and mitigate the risk over-fitting. The models to be developed in REALM will provide accurate reaction profiles for new reactants in real-time, thus circumventing the need for expensive Quantum Mechanical (QM) calculations and/or the development of novel catalysts by trial-and-error. This will translate into a predictive computational tool which allows chemists to easily optimise reaction conditions, develop new catalysts, or plan new synthetic pathways.
In particular, the ML methods to be developed in REALM will be tested towards the exploration of the chemical space concerning the reduction of nitro-arenes into amino-arenes, relevant from the point of view of both synthesis and the environment.