Correlating the partitioning of organic molecules between water and [MeoeMPyrr]+ [FAP]- through machine learning
Published in Computer Aided Chemical Engineering, 2024
Lipophilicity is one of many parameters involved in the biological activity of drugs. It is assessed by defining the partitioning of a molecule between an organic (i.e. octanol) and a water phase. Nevertheless, octanol is too simple to encode all of the complicated interactions seen in ionic liquids. Moreover, the experimental determination of logP in specific ionic liquid/water systems (logPIL/W) is an arduous and resource-intensive task. Machine learning and hybrid modelling techniques have emerged as essential tools in chemical engineering, providing innovative solutions to complicated physicochemical problems. This study proposes a hybrid model correlating the partitioning of organic molecules in octanol/water with the partitioning in [MeoeMPyrr]+[FAP]-/water systems. The hybrid model is formed by a first principle and a data-driven part. The first is represented by a group contribution model, the latter is represented by an ensemble of 5 Multilayer Perceptron (MLP) models. The model structure with the highest accuracy and generalization properties is searched with 5-fold cross-validation, using hyperparameter optimization.. The prediction capabilities have been evaluated through various metrics, namely the coefficient of determination (R2 = 0.93), the Mean Squared Error (MSE = 1.8·40-2) and the Mean Absolute Percentage Error (MAPE = 25%), showing quite good accuracy.
Recommended citation: Esposito, F., Di Caprio, U., Vermeire, F., Leblebici, M.E. (2024). Correlating the partitioning of organic molecules between water and [MeoeMPyrr]+ [FAP]- through machine learning. Computer Aided Chemical Engineering, 53, 2959-2964