A fuzzy system for modeling the structure-activity relationships in presence of uncertainties

This study deals with the development of a computer model to estimate the biological activity of a chemical compound by evaluating its molecular structure. There exist uncertainties in a structure-activity modeling problem regarding the numerical representation of the structure via selected molecular descriptors, optimal structure of the model, robustness of the model identification criterion, and presence of noise in experimental activity data used to identify the model. Our approach to deal with the uncertainties in modeling problem consists of following steps: a) The uncertainties in the modeling problem are filtered out using a fuzzy filter. b) Stochastic modeling of the uncertainties using finite mixture models and utilizing this information regarding uncertainties for adapting the structure-activity models to approximate the uncertain relationships. The method is illustrated via considering a case study of modeling the activity of 3-phosphoinositide-dependent protein kinase-1 (PDK1) inhibitors.