Abstract
De novo molecular design aims to propose molecules exhibiting desired properties and/or activities, as constructed from scratch. Although this approach opposes the widely used virtual screening (VS), the same criteria should be applied, such as ones based on substructure filters, and quantitative structure-property relationship (QSPR) and quantitative structure-activity relationship (QSAR) regression models. QSPR/QSAR, which enables us to predict properties/activities by making use of experimental data, are widely used in academia as well as in industry. Herewith, we present a novel chemical structure generation system by combining fragments whose final chemical structures satisfy the aforementioned criteria. Using inverse analysis, QSPR/QSAR models determine a specific region in chemical space corresponding to a set of desired values by a designer. Chemical structures are generated by combining ring systems, as well as atom fragments, in every possible way until violating the upper bounds of that region. We also show the results of inverse-QSAR analysis for the human Alpha-2A adrenergic receptor. This suggests that our system has features preferable to VS-like methods in terms of the number of generated structures.
