You are here:
Low Dimensional Chemical Spaces Via Manifold Learning
Citation:
Charest, N., C. Ramsland, C. Lowe, T. Martin, AND A. Williams. Low Dimensional Chemical Spaces Via Manifold Learning. Fall ACS, Chicago, IL, August 21 - 25, 2022. https://doi.org/10.23645/epacomptox.20503365
Impact/Purpose:
N/A
Description:
Chemical space is the abstract term for any mathematical description of chemical structures and their relationships to one another. Applied developments of chemical space often involve the selection of chemically meaningful ‘descriptors’ which act as coordinates for a structure’s position in the chemical space, with subsequent analysis performed based on this coordinate system to determine the chemical’s relationships and properties. To facilitate visualization and to avoid the infamous ‘curse of dimensionality’, in which mathematical systems become increasingly unpredictable with each additional degree of independence, it is desirable to formulate meaningful chemical spaces using relatively few coordinates. We explore manifold learning as a potential means of condensing the information from highly parameterized descriptions of chemical space into visualizable dimensionality. The relevance of these low dimensional chemical spaces to issues of modeling such as coverage, model trends and prediction consistency are explored. These considerations are made for QSA/PR (Quantitative Structure Activity/Property Relationship) models across multiple analytical, physicochemical and toxicological endpoints. This abstract does not necessarily represent the views or policies of the U.S. Environmental Protection Agency.
URLs/Downloads:
DOI: Low Dimensional Chemical Spaces Via Manifold Learning
MANIFOLD_EMBEDDING WITH DISCLAIMER.PDF (PDF, NA pp, 383.849 KB, about PDF)