A comparison among generalized linear models family for drug solubility prediction based on molecular structure

Abstract

In all the property of molecular which can significantly influence a compound's biological activity, in the initial stages of drug discovery, aqueous solubility is likely a highlight amongst the most central and consideration. Two important elements that influence a medication's oral bioavailability are aqueous solubility and membrane permeability. In view of the significance of aqueous solubility, a lot of attempts have been used to creating dependable models to forecast this physiochemical property. Though some advancement has been made and a lot of models have been developed, it is presumed that exact and dependable aqueous models focused to anticipate the dissolvability of drug-like molecules have not happened. Based on the principle of molecular similarity which states that structurally similar chemical compounds will very likely have similar aqueous solubility, this research is conducted to construct a predictive model for identifying the drug solubility by molecular structure from ESOL database. The model developed the relation of aqueous solubility and their structure-based descriptors. The family of generalized linear model was successfully built and compared each type of model with the others. Finally, these models show the high predictive performances on both training and test set, suggesting that these models are reliable in predicting the relationship between the structures of drug compounds and their aqueous solubility. Keywords: Generalized linear model Aqueous solubility Drug discovery Molecular structure Machine Learning