Improving the predictive ability of drug solubility in in-vitro intestinal fluids through hybrid modeling strategies

The equilibrium solubility of a drug in the gastrointestinal tract is a key parameter for assessing the availability of the drug through the human body. The complexity of the extraction and manipulation of human intestinal fluids determines that few experiments can be carried out on in vivo patients and, for this reason, measurements of solubility are generated in vitro on Simulated Intestinal Fluids (SIFs). The main objective of this Thesis is to accelerate the identification of poorly soluble drugs during the phase of drug development, aiding at the same time the reduction of time and resources utilized for experimentation. The proposed strategy utilizes Gaussian Process regression, which increase the accuracy of predictions of drug equilibrium solubility from in-vitro experimental measurements. The proposed models account for food effects, and, at the same time, accurately describe inter- and intra-subject variability by implementing the GP regression model of intestinal solubility. These modeling strategies can be implemented within physiologically based pharmacokinetic models enabling to increase the accuracy of drug availability. Furthermore, the proposed strategy improves the extraction of information and reduces the experimental burden because of its ability to perform predictions of new unseen intestinal compositions. The proposed methodology demonstrated to be a general procedure which can be applied to different classes of drugs. In particular, in this Thesis it was applied to three case studies (i.e., different drugs): an Active Pharmaceutical Ingredient (from in vitro experimentation of Stamatopoulos et al. (2023)), Felodipine for dealing with arterial hypertension and Fenofibrate for high cholesterol treatment.