Application of machine learning in prediction of hydrotrope-enhanced solubilisation of indomethacin

Abstract

Systematic in-vitro studies have been conducted to determine the ability of a range of 10 potential hydrotropes to improve the apparent aqueous solubility of the poorly water soluble drug, indomethacin. Solubilisation of the drug in the presence of the hydrotropes was determined experimentally using high-performance liquid chromatography (HPLC) with ultraviolet (UV) detection. These experimental data, together with various known and computed physicochemical properties of the hydrotropes were thereafter used in silico to train an artificial neural network (ANN) to allow for predictions of indomethacin solubilisation. The trained ANN was found to give highly accurate predictions of indomethacin solubilisation in the presence of hydrotropes and was thus shown to provide a valuable means by which hydrotrope efficacy could be screened computationally. Interrogation of the network connection weights afforded a quantitative assessment of the relative importance of the various hydrotrope physicochemical properties in determining the extent of the enhancement in indomethacin solubilisation. It is concluded that in-silico screening of drug/hydrotrope systems using artificial neural networks offers significant potential to reduce the need for extensive laboratory testing of these systems, and could thus provide an economy in terms of reduced costs and time in drug formulation development.