Global structural similarity in chemical compounds

Quantitative Structure-Activity Relationship studies are a field of work in biochemistry that seeks to provide an empirical, semi-empirical or theoretical basis for estimating the physiochemical properties of chemical compounds. Knowledge of these properties allows scientists to predict aspects of a compound's behavior without extensive experimentation beforehand, making these methods valuable for several applicative domains. We approach structure-activity relationships from a computational prospective, by using a simple structural encoding method. We produce an algorithm that compares the resulting graph-like representations of chemical compounds, and investigate whether global structure, rather than "local" features of molecules, can be used to infer bio-chemical activity. Our results confirm that we can do this for thermodynamic properties with accuracy comparable to well established methods, and that global structural similarity between compounds seem to hold for more intricate, biological behaviors.