Assessment of physicochemical properties of materials using a model-based system

This paper describes the basic ideas underlying the design of a computational modeling system that has been successfully employed for the assessment of the mechanical behavior of a class of polymers (different grades and concentrations of Sodium Carboxymethylcellulose, NaCMC), candidate for use in Pharmacology as drug delivery systems. Given in input creep data obtained by performing standard rheological tests on a sample, the system automatically provides in output a quantitative model (ordinary differential equation) of the mechanical behavior of the tested material. With the aim of emulating the skillful human expert in tackling the modeling problem, the computational system embodies a variety of qualitative and quantitative techniques, which allow it to make the most of the available knowledge. The main steps in the modeling process, given the experimental data and a previously generated exhaustive library of models of ideal visco-elastic materials, are: 1) intelligent data analysis, 2) structural and 3) quantitative identification. The former two, which strongly rely on qualitative reasoning techniques as well as on domain specific knowledge, consist in selecting within the library a subset of models that match the qualitative profiles abstracted from the observations. This task mimics the expert as he visually inspects the data, recognizes physical features from characteristic geometric patterns, and makes an initial guess for one or more models adequate to explain the observed features. Quantitative identification then refines the model search process within the plausible model subset through an optimization loop, which ends up selecting the quantitative model that ensures a good fit to data while meeting the parsimony principle. The role of QR within the modeling process is crucial as it makes it more efficient by reducing the model search space and driving the numerical methods, and, even most importantly, as it makes it more robust by guaranteeing that the resulting model provides an optimal balance between physical accuracy and goodness of fit. As a consequence, the number and values of the model parameters truly hold a physical meaning, providing a deeper insight into the structure of the observed system. In Pharmacology, the availability of an automated modeling tool can favour a model-based approach to the investigation of physico-chemical properties of polymers, which has been so far carried out by means of purely experimental studies. In fact, the large body of interdisciplinary expert knowledge required of researchers to formulate models has often been a deterrent from using them. On the other hand, models allow researchers to explore more deeply how material complexity changes in accordance with planneded modifications in its chemical structure. As regards the specific application, the model-based approach has allowed us to shed light onto the polymer-mucin interactions involved in mucoadhesion, which is one of the chemico-physical features that directly influence drug bioavailability. More precisely, solutions of NaCMC polymers and polymer-mucin mixtures have been tested at different concentrations. The experimental data have been processed through the computational framework, and accurate models for each data set have been obtained. The model order and the values of the parameters respectively indicate the number of structural units within the polymeric network and their strength, and therefore characterize the material complexity. By comparing the values obtained for the different samples, we could explain the strengthening of the mucoadhesive interface in terms of the structural variations that take place in the polymeric network: the interface strengthening could be caused either by the establishment of new types of linkages, or by an increased rigidity of pre-existing ones.