(72a) Machine Learning and COSMO-RS Assisted to Predict the Solvatochromic Parameters of Deep Eutectic Solvents for Lignocellulosic Biorefinery

Ionic liquids (ILs) and deep eutectic solvents (DES), emerging in recent years, are green solvents to replace the traditional organic solvent for lignocellulosic pretreatment and biorefinery due to their excellent performance: low vapor pressure, good stability, recyclability, structure tunability, ...etc¹. Rapid design and discovery of suitable ILs or DES for lignocellulosic pretreatment demands knowing the solvation properties and solvatochromic parameters. Kamlet and Taft put forward solvatochromic parameters to describe the polarity. They are hydrogen bond acidity (α), hydrogen bond basicity (β), and dipolarity/polarizability effects (Ï€*). Solvatochromic parameters can interpret the relationship between DESs’ properties and the efficiency of lignin fractionation, as well as the prediction of lignin solubility. Computational software has been reported and proven useful tools to predict the solvatochromic parameters of various ILs and DES². Conductor-like Screening Model for Real Solvents (COSMO-RS) was used to predict the thermodynamic properties of fluids based on the quantum chemistry. Although there have been some studies on the prediction of solvatochromic parameters of ILs or DES based on COSMO-RS calculation, it is still a huge challenge to build a reliable computational model with sufficient data to predict the three solvation parameters.

In this work, for the first time, three solvation parameters of ILs and DESs are predicted together by machine learning and COSMO-RS calculations. The Kamlet−Taft solvatochromic parameters experimental data of 175 common organic solvents, 214 ILs, and 110 DESs were collected, and the COSMO-RS descriptor (Sσ-profiles descriptor) of each compound was defined and calculated, neural network fitting app of MATLAB was used to fit the data. A reliable model to predict the solvation parameters of ILs and DESs was developed in the final. Experimental tests of the solvation parameters of synthesized ILs and DES further verify the reliability of the machine-learning model. A novel DES with higher α and β was designed and synthesized based on the neural network model. The novel DES showed high corn stover pretreatment efficiency and lignin removal rate, the pretreated corn stover was easy to enzymolysis and produce polyhydroxyalkanoates by engineered Pseudomonas putida KT2440³. Overall, improving lignocellulosic waste valorization to bioproducts by employing novel DES pretreatments provided a sustainable biorefinery process. Machine learning and sufficient experimental data ensure the reliability of the model established, which not only predicts and designs the DES for lignocellulosic biorefinery but also has great potential application in the design and selection of subsequent functional ILs and DES.

1. Yuan, J. S.; Pavlovich, M. J.; Ragauskas, A. J.; Han, B. X., Biotechnology for a sustainable future: biomass and beyond. Trends Biotechnol 2022, 40 (12), 1395-1398.
2. Wojeicchowski, J. P.; Abranches, D. O.; Ferreira, A. M.; Mafra, M. R.; Coutinho, J. A. P., Using COSMO-RS to Predict Solvatochromic Parameters for Deep Eutectic Solvents. Acs Sustainable Chemistry & Engineering 2021, 9 (30), 10240-10249.
3. Liu, Z. H.; Hao, N. J.; Wang, Y. Y.; Dou, C.; Lin, F. R.; Shen, R. C.; Bura, R.; Hodge, D. B.; Dale, B. E.; Ragauskas, A. J.; Yang, B.; Yuan, J. S., Transforming biorefinery designs with 'Plug-In Processes of Lignin' to enable economic waste valorization. Nat Commun 2021, 12 (1).