The identification of proper molecular descriptors is essential for the virtual screening of PFASs and mapping of the chemical-property space. The descriptors may be physicochemical (hydrophobic, steric, or electronic), structural (based on frequency of occurrence of a substructure), topological, electronic (based on molecular orbital calculations), geometric (based on a molecular surface area calculation), or simple indicator parameters (dummy variables). There exist a variety of molecule descriptors which capture and magnify distinct aspects of chemical structures. However, many of these descriptors are dependent upon one another and sometimes even capture the same information. Thus, the selection of the appropriate molecular descriptors to incorporate into a certain analysis is a nontrivial task. Recent works have shown the incorporation of relatively simple descriptions of the molecular structure have shown limited success for the prediction of bond dissociation energy using a variety of machine learning algorithms. In such works, the “spatial” location of particular atoms of interest are represented in a series of binary bits that indicate the presence or absence of particular substructures in the molecules. These “Fingerprints” can be used as descriptors; however, the appropriate substructures must be carefully selected for a given application. We note the characterization techniques may oversimplify the structural characterization of these molecules into atom-pair relations, subsequently, much of the structural information may be lost during this process. Thus, in this work a more rigorous methodology for the characterization of the molecular structure will be explored using voxelization techniques. The explored voxelization techniques will then be used for the predictions of bond dissociation energy. It is anticipated the techniques deployed in this work will have widespread applicability in establishing structure-activity/structure-property linkages.

Exploration of machine learning approaches for feature engineering of the PFAS molecular structure and correlation to molecular properties

ACS Spring Meeting 2021

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