A problem of separation of trivalent f-elements is one of key problems of spent fuel reprocessing as well as processing of lanthanide-containing mineral resources. Chemical similarity leads to development of long cascade separation equipment with low efficiency on each step. Development of new selective ligands may allow us to simplify separation scheme and increase process efficiency.
"Deep learning" is one of popular trends in science and technology nowadays. It has already proved its chemical usability in drug design area and continues spreading over other areas of chemistry.
Here we present results of development of neural net models that were trained to predict complex stability constants for a number of trivalent f-elements. While requesting a lot of time and computational resources for training, such models are easy in use and quite fast for prediction purposes. We reached determination coefficient (R^2) values up to 90~95% that allow us to determine perspective in terms of separation ligands at a design stage.