Molecular Graph Modularity as a Descriptor for Property EstimationApplication to the Viscosity of Biomass-Derived Molecules

The
diversity of biomass and its conversion processes produce a
diverse pool of functional molecules. For many of these molecules,
property data have not been measured yet and need to be estimated
to determine suitability to a particular application. For example,
viscosity is a key property in the development of green solvents,
fuel additives, and biofuels. This paper proposes the use of modularity
as a molecular descriptor combined with functional group counts to
estimate molecular properties using neural network models. The modularity
of molecules was determined from graph representations using community
detection algorithms. The potential for this approach was demonstrated
for the modeling of viscosity at 25 °C and applied to biomass-derived
molecules. The model performances showed that including modularity
contributed to more accurate estimations than viscosity models existing
in the literature. Furthermore, modularity on its own can be useful
to estimate viscosity for n-alkanes, esters, isoalkanes,
aldehydes, aromatics, and cycloalkanes. This was due to the capacity
of modularity to capture the structural features of the molecules
employed in the data set. As such, modularity exhibited its tremendous
potential to be exploited for property estimation, supporting the
screening, rational design, and engineering of green chemicals derived
from biomass.