A neural network model for rapid prediction of analyte focusing in isotachophoresis

We present the development and demonstration of a neural network (NN) model for fast and accurate prediction of whether or not a chosen analyte is focused by an isotachophoresis (ITP) buffer system. The NN model is useful in the rapid evaluation of possible ITP chemistries applicable to analytes of interest. We trained and tested the NN model for univalent species based on extensive data sets of over 10,000 anionic and 10,000 cationic ITP simulations. The NN model uses as inputs the mobilities and the acid dissociation constants of leading electrolyte ion, trailing electrolyte ion, counterion, and a single analyte as well as the leading-to-counterion concentration ratio of the leading zone. The output then indicates whether the chosen electrolyte system yields stable ITP focusing of the analyte. The prediction accuracy of the NN model is over 97.7%. We demonstrate the applicability of the NN by validating its predictions with reported experimental data for anionic and cationic ITP. We have packaged the NN model in a free, web-based application named IONN (isotachophoresis on neural network), which can be used to rapidly screen ITP electrolyte systems.