Matched Molecular Pairs (MMPs) have become a standard tool in QSAR analysis over the last decades. Their usefulness was proven in a number of publications in the areas of ADME, bioisosterism, aqueous solubility, logD, and several more. At the core of MMP analysis and QSAR methods in general lies the assumption that the effect of chemical substitution can be generalized. This assumption holds true as long as the environment of the location of the substitution is homogeneous. Heterogeneous environments like the varying amino acid composition of binding pockets of different targets are leading to false property predictions. One solution would be to move from purely ligand-based descriptions to models that include the information of the protein target. The ever increasing amount of published protein-ligand complex data, better algorithms for superimposition of two structures and faster computing units renders working with three-dimensional (3D) data worthwhile. 3D descriptors extract information about a system impossible to get with purely ligand-based methods. This additional information allows for a better abstraction of chemical interaction models and thorough analysis of the impact of transformations applied to the ligand. Activity cliffs for example are easily analyzed with information gained from 3D models. Here, we present a novel workflow of an automatically generated dataset of 3D matched molecular pairs (3DMMPs). The protein-ligand structures are extracted from the PDB database and subjected to a rigorous assessment process. Parameters like resolution, DPI, Model significance, EDIA score, Lipinski rules of five - just to name a few - are calculated and each protein structure is classified based upon these numbers. Items passing user defined thresholds are pooled, MMPs generated and activity values are extracted from a user specified source, usually ChEMBL or PubChem. Form this point on two different sets of data are available. In decreasing level of quality: a) 3DMMPs purely from crystal structures and b) 3DMMPs where one ligand is virtual. Virtual ligands are those where an activity value can be found but no crystal structure exists. The workflow presented has been combined into a single software solution enabling researchers to easily assemble a data set of 3DMMPs, tailored to their focus and the quality level needed. The usefulness of this approach will be exemplified by analyzing various interesting MMPs automatically assembled from PDB and ChEMBL.

Automatic generation of high quality 3D matched molecular pairs

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.