Comparison between artificial neural network and density-based models for the correlation of the solubility of some pharmaceutical compounds in supercritical carbon dioxide

摘要

This work compares Artificial Neural Networks (ANN) to four density based semi empirical modelling of the solubility of seven non-steroidal anti-inflammatory (NSAID), two anti-Cancer and two anti-AIDS drugs in supercritical carbon dioxide (scCO_2). Experimental literature data for the eleven drugs were used for training (152 data points) and validation (75 data points) of the ANN model. The model has five inputs (two intensive state variables and three pure drug properties) and one output (solubility of solid drug in scCO_2 in mole fraction). Statistical analysis of the predictability of the neural networks model shows excellent agreement with experimental data. Furthermore, the comparison in terms of average absolute relative deviation (AARD) between the ANN predicted results for each binary for the whole temperature and pressure range and results predicted by four density based models (Chrastil, Kumar and Johnston, Bartle et al., and Mendez-Santiago and Teja) show that the ANN model correlates far better the solubility of the eleven solid drugs in scCO_2.