Skip to main content
ARS Home » Southeast Area » New Orleans, Louisiana » Southern Regional Research Center » Commodity Utilization Research » Research » Publications at this Location » Publication #384568

Research Project: Improved Conversion of Sugar Crops into Food, Biofuels, Biochemicals, and Bioproducts

Location: Commodity Utilization Research

Title: Estimation of Hansen solubility parameters with regularized regression for biomass conversion products: An application of adaptable group contribution

item Terrell, Evan

Submitted to: Chemical Engineering Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/4/2021
Publication Date: 2/2/2022
Citation: Terrell, E. 2022. Estimation of Hansen solubility parameters with regularized regression for biomass conversion products: An application of adaptable group contribution. Chemical Engineering Science. 248:117184.

Interpretive Summary: Creating new products from renewable, biomass-based resources presents several challenges related to their characterization in the laboratory. Many of the technologies that currently exist for the production of fuels and/or chemicals from biomass give us complicated mixtures of products made up of a variety of molecules. Because of this complexity and the general lack of purity resulting from a biomass conversion process, designing experiments to measure interesting properties of these products is very difficult. This paper in particular is interested in the solubility properties of biomass conversion products. A new and improved method to estimate these solubility properties in a computational way is introduced here. The goal of this work is to allow researchers to adapt this computational approach to their own systems and datasets, allowing for more flexibility in the application of the proposed method. There is also good opportunity to extend the analysis presented here to the computation of other interesting properties, not limited only to solubility. By using computers to understand our systems in a hypothetical way at low cost, more expensive experimental work can be better targeted at the things that are most interesting or show the most promise for further engineering development.

Technical Abstract: There is increasing growth in research of the development, upgrading and valorization of forestry and agriculture-derived products for applications in fuels, chemicals, and materials. Many biochemical and thermochemical biomass conversion technologies yield unique products for which laboratory characterization of structure and properties is still an ongoing challenge. Historically, popular approaches to estimate properties (e.g., boiling point, critical point, solubility parameters) from structure have relied on available, published group contribution methods. In this study, the Hansen solubility parameters of biomass-derived compounds are estimated using regularized regression as a platform for adaptable group contribution. A training set of small molecules and a set of biomass conversion molecules (heavy pyrolysis oil products and proposed bioprivileged molecules) are parameterized using simple contributing groups. Regularized regression is then applied as a tool to reduce model features and complexity in estimating the solubility parameters. Using regularized regression (specifically, least absolute shrinkage and selection operator) allows for the user to develop their own contributing groups at any sufficient level of complexity, which are then down-selected to only those which are most important, while avoiding overfitting. The linear model itself is built upon readily available experimental data and needs no proprietary software, using only python and its popular science/data analysis libraries. The model also shows good agreement with other recently published work designed for "pencil and paper" estimation of Hansen solubility parameters. The combination of regularized regression with adaptable group contribution has potential for applications in the prediction of other molecular properties for which group contribution methods are commonly employed.