Dresden 2017 – wissenschaftliches Programm
CPP 16.2: Poster
Montag, 20. März 2017, 18:30–21:00, P1C
Systematic Reduction of Chemical Compound Space using Coarse-Graining and Clustering Algorithms — •Kiran Kanekal, Kurt Kremer, and Tristan Bereau — Max Planck Institute for Polymer Research
The size of chemical compound space is prohibitively large for the fast generation of hypersurfaces that define structure-property relationships, which are necessary for implementing inverse molecular design. In this work, we use a dataset consisting of all small organic molecules with up to 8 heavy atoms (i.e. excluding hydrogen atoms) as an initial proxy for the entire chemical compound space. This relatively small slice of the total space consists of ~100,000 molecules. We demonstrate a reduction of this space in two steps. First, we apply the AutoMartini algorithm developed by Bereau and Kremer to systematically determine a coarse-grained representation for each molecule in the dataset. We subsequently cluster molecules by using similarity measures corresponding to specific chemical and structural descriptors. We then assess the extent to which the initial chemical compound space was reduced and whether the diversity of the space is sufficiently reflected in its coarse-grained counterpart. The resolution of the coarse-grained space can then be tuned either by introducing new bead types or by reducing the number of atoms assigned to a bead. This framework will provide a means for efficient high throughput sampling of chemical compound space (as highlighted by Menichetti and coworkers in their work).