Exploring chemical space for drug discovery using the chemical universe database

Abstract

Herein we review our recent efforts in searching for bioactive ligands by enumeration and virtual screening of the unknown chemical space of small molecules. Enumeration from first principles shows that almost all small molecules (>99.9%) have never been synthesized and are still available to be prepared and tested. We discuss open access sources of molecules, the classification and representation of chemical space using molecular quantum numbers (MQN), its exhaustive enumeration in form of the chemical universe generated databases (GDB), and examples of using these databases for prospective drug discovery. MQN-searchable GDB, PubChem, and DrugBank are freely accessible at www.gdb.unibe.ch.

Figure 1

Color-coded MQN-map of the PubChem chemical space (19.2 million structures) as the (PC2,PC3)-plane, marked as the horizontal and vertical axes starting from the (0,0) coordinate where hydrogen is located. The values corresponding to each color are indicated on the maps. PC2, PC3 refer to the 2nd and 3rd principal component, respectively, in the PCA of the MQN data for PubChem. (A) Average number of non-hydrogen atoms per molecule. (B) Average fraction of cyclic atoms per molecule. (C) Average fraction of H-bond acceptor atoms per molecule. (D) Compound categories including computationally enumerated molecules (up to hac = 500) for each category. Ro3 are Congreve’s “rule of 3” molecules, and Ro5 are Lipinski’s “rule of 5” molecules.

Figure 2

Overview of the chemical universe database GDB-13 containing 977 million structures up to 13 atoms of C, N, O, Cl, S. (A–C) Color-coded MQN-map of the (PC1,PC2) plane. PC1 (horizontal dimension) and PC2 (vertical dimension) refer to the 1st respectively 2nd principal component in the PCA of the MQN data for GDB-13. (D) Size of the GDB database, its 43.7 M subset, and PubChem as a function of molecular size.

Figure 3

NMDA glycine site and GLT-1 inhibitors identified from GDB-11 by virtual screening, synthesis and testing. Activities were determined by electrophysiology (NMDA glycine site) or by radioactive ligand uptake inhibition (GLT-1) for the human receptors expressed in Xenopus oocytes.

Figure 4

Discovery of α7 nAChR inhibitors by fragment-based diversification of known ligands using GDB-11.

Figure 5

Discovery of α7 nAChR inhibitors by nearest neighbor searching in the MQN-space of GDB-13.