|Sunghwan Kim|| at 11:00
Application of Molecular Shape Quadrupoles to Develop a Prescreening Scheme for PubChem 3D Neighboring Computation
When molecules are structurally similar to each other, they are likely to have similar biological and physicochemical properties. This so-called "similarity principle" is an important concept in ligand-based drug design often applied to find potential drug candidates at the initial stages of drug discovery and development. PubChem provides a pre-computed 3D neighboring relationship (good for finding chemicals with similar 3D shape and feature orientation) in addition to a so-called 2D similarity relationship (good for finding similar chemical analogs). The PubChem3D neighboring computation involves finding the maximal shape overlap between two static compound conformations using the Shape-Tanimoto (ST) similarity score, considered to be an accurate measure for 3D molecular shape comparison. However, because the shape overlay optimization computation is the most time-consuming step in the PubChem3D neighboring computation, it is highly desirable to develop a prescreening scheme that enables us to avoid as many optimizations as is possible. In the present study, molecular shape quadrupole moments, conceptually equivalent to the length, width, and height of a molecule, were employed to devise prescreening filters. In a test run with ~13 billion pairs of molecules, the new filters were found to be highly efficient, with minimal CPU overhead and able to filter out ~5 billion non-neighbor pairs with effectively no 3D neighbor loss. This presentation will detail the development of these filters.