The Theoretical Chemistry Research Group at TTK participated to an initiative about introducing their research activities, to secondary school students interested in chemistry. The video is in Hungarian with English subtitles and our project is shown between minutes 1:05 and 3:35″, here the English text of what our colleagues saying about the project:
Our EU CompBat project has reached its final stage and our newest results will be presented next week in Uppsala. Flora will tell you more about this.
An international team involving six different research groups is engaged in the development of new batteries. One of the key aspects of this project is the discovery of simple organic or inorganic molecules that have advantageous electrochemical properties. One of the most important properties here is the redox potential, which describes how favorable it is for the molecule to gain or lose an
>> FLORA: We wish to develop new water-based redox flow batteries and the task of our group is to design a computational screening protocol. We have created a molecular database of more than 8000 molecules via quantum chemical calculations. It contains information about the willingness of these molecules to gain an electron. The database was used to train five different deep learning models. These models take the structures of the molecules and their redox potentials and they infer the relationship between the two properties. With the learned correlation, the models now can be used to predict the redox potentials for molecules not included in our database.
>> P. IMRE: If we had to perform 1000 experiments to find a solution, the computations can identify not only those 900 experiments that are not useful to be performed, but also those ten that will likely lead to success.
>> B. IMRE: You assumed that these reactions take place in aqueous solutions. However, if I introduce an arbitrary molecule to the model, it may not be water soluble.
>> FLORA: The water solubility of the compounds can be estimated via computations, but we will not be able to get accurate values. Here, the goal was simply to include as many molecules as possible for the machine learning algorithms.
>> GERGO: But we can expect from an organic chemist to tell us whether the molecule is water soluble or not, can’t we?