Predicting the Fine Particle Fraction of Dry Powder Inhalers Using Artificial Neural Networks.

Dry powder inhalers are increasingly popular for delivering drugs to the lungs for the treatment of respiratory diseases, but are complex products with multivariate performance determinants. Heuristic product development guided by in vitro aerosol performance testing is a costly and time-consuming process. This study investigated the feasibility of using artificial neural networks (ANNs) to predict fine particle fraction (FPF) based on formulation device variables. Thirty-one ANN architectures were evaluated for their ability to predict experimentally determined FPF for a self-consistent dataset containing salmeterol xinafoate and salbutamol sulfate dry powder inhalers (237 experimental observations). Principal component analysis was used to identify inputs that significantly affected FPF. Orthogonal arrays (OAs) were used to design ANN architectures, optimized using the Taguchi method. The primary OA ANN r2 values ranged between 0.46 and 0.90 and the secondary OA increased the r2 values (0.53-0.93). The optimum ANN (9-4-1 architecture, average r2 0.92 ± 0.02) included active pharmaceutical ingredient, formulation, and device inputs identified by principal component analysis, which reflected the recognized importance and interdependency of these factors for orally inhaled product performance. The Taguchi method was effective at identifying successful architecture with the potential for development as a useful generic inhaler ANN model, although this would require much larger datasets and more variable inputs.

In silico prediction of aqueous solubility using simple QSPR models: the importance of phenol and phenol-like moieties.

Recently the authors published a robust QSPR model of aqueous solubility which exploited the computationally derived molecular descriptor topographical polar surface area (TPSA) alongside experimentally determined melting point and logP. This model (the "TPSA model") is able to accurately predict to within ± one log unit the aqueous solubility of 87% of the compounds in a chemically diverse data set of 1265 molecules. This is comparable to results achieved for established models of aqueous solubility e.g. ESOL (79%) and the General Solubility Equation (81%). Hierarchical clustering of this data set according to chemical similarity shows that a significant number of molecules with phenolic and/or phenol-like moieties are poorly predicted by these equations. Modification of the TPSA model to additionally incorporate a descriptor pertaining to a simple count of phenol and phenol-like moieties improves the predictive ability within ± one log unit to 89% for the full data set (1265 compounds -8.48 < logS < 1.58) and 82% for a reduced data set (1160 compounds 6.00 < logS < 0.00) which excludes compounds at the sparsely populated extremities of the data range. This improvement can be rationalized as the additional descriptor in the model acting as a correction factor which acknowledges the effect of phenolic substituents on the electronic characteristics of aromatic molecules i.e. the generally positive contribution to aqueous solubility made by phenolic moieties.

Revisiting the general solubility equation: in silico prediction of aqueous solubility incorporating the effect of topographical polar surface area.

The General Solubility Equation (GSE) is a QSPR model based on the melting point and log P of a chemical substance. It is used to predict the aqueous solubility of nonionizable chemical compounds. However, its reliance on experimentally derived descriptors, particularly melting point, limits its applicability to virtual compounds. The studies presented show that the GSE is able to predict, to within 1 log unit, the experimental aqueous solubility (log S) for 81% of the compounds in a data set of 1265 diverse chemical structures (-8.48 < log S < 1.58). However, the predictive ability of the GSE is reduced to 75% when applied to a subset of the data (1160 compounds -6.00 < log S < 0.00), which discounts those compounds occupying the sparsely populated regions of data space. This highlights how sparsely populated extremities of data sets can significantly skew results for linear regression-based models. Replacing the melting point descriptor of the GSE with a descriptor which accounts for topographical polar surface area (TPSA) produces a model of comparable quality to the GSE (the solubility of 81% of compounds in the full data set predicted accurately). As such, we propose an alternative simple model for predicting aqueous solubility which replaces the melting point descriptor of the GSE with TPSA and hence can be applied to virtual compounds. In addition, incorporating TPSA into the GSE in addition to log P and melting point gives a three descriptor model that improves accurate prediction of aqueous solubility over the GSE by 5.1% for the full and 6.6% for the reduced data set, respectively.