How Far Do Kamlet–Taft Parameters and pH Predict Delignification by Deep Eutectic Solvents?

Deep eutectic solvents (DESs) are tunable, low-toxicity media for lignocellulose fractionation, but their selection for delignification still relies on empirical screening. Kamlet–Taft (KT) solvatochromic parameters (hydrogen-bond acidity α, basicity β, and dipolarity/polarizability π*) and pH are the descriptors most often correlated with lignin solubility, yet how far they predict delignification of whole biomass has not been tested critically. Here we report measured KT parameters and apparent pH (pHapp) for 56 DES formulations spanning carboxylic-acid, polyol, metal-salt, amide, alkanolamine, and amino-acid families, together with the softwood lignin removal obtained after a sequential enzymatic–DES fractionation. Four descriptor-to-performance models were compared under identical leave-one-out, repeated k-fold, and grouped (leave-one-cluster-out and leave-one-family-out) cross-validation: a linear correlation, a 15-term second-order polynomial, a random forest (RF), and a Gaussian-process (GP) regressor with automatic relevance determination. All models fit the training set (in-sample R2 up to 0.89 for the RF) but generalize weakly: leave-one-out R2 is 0.10–0.13 for the GP and the RF, the polynomial is overparameterized and generalizes worse than the training mean, and no model predicts a chemical family that is absent from training better than the training mean. Gaussian-process relevance and ablation and random-forest permutation importance rank pHapp as the most informative descriptor, followed by α, whereas β and π* carry little consistent information, and the pKa of the hydrogen-bond donor carries the same information as pHapp; these results point to solvent acid strength, rather than hydrogen-bond donation alone, as the property most closely associated with lignin removal from enzymatically pretreated softwood. The GP provides calibrated predictive uncertainty: 96% of leave-one-out predictions and all four prospectively synthesized blind-test DESs fall within its 95% credible intervals. The results delimit what solvatochromic descriptors can and cannot predict for whole-biomass delignification and provide an uncertainty-aware screening tool and an open data set for DES design.

Authors

Institutions

Publication Details

Journal
ACS Omega
Published
2026-11-18
DOI
https://doi.org/10.1021/acsomega.6c10660
Primary Topic
Biofuel production and bioconversion
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

How Far Do Kamlet–Taft Parameters and pH Predict Delignification by Deep Eutectic Solvents?

Guadalupe Pérez-Durán, Mert Atilhan, Santiago Aparício, Swati Sood et al.
3 links
ACS Omega
Biofuel production and bioconversion
article

How Far Do Kamlet–Taft Parameters and pH Predict Delignification by Deep Eutectic Solvents?

Guadalupe Pérez-Durán, Mert Atilhan, Santiago Aparício, Swati Sood, Gustavo Arturo Iglesias-Silva, Supti Paul, Roman Brunecky, Alberto Gutiérrez, Kecheng Li
article en

Abstract

Deep eutectic solvents (DESs) are tunable, low-toxicity media for lignocellulose fractionation, but their selection for delignification still relies on empirical screening. Kamlet–Taft (KT) solvatochromic parameters (hydrogen-bond acidity α, basicity β, and dipolarity/polarizability π*) and pH are the descriptors most often correlated with lignin solubility, yet how far they predict delignification of whole biomass has not been tested critically. Here we report measured KT parameters and apparent pH (pHapp) for 56 DES formulations spanning carboxylic-acid, polyol, metal-salt, amide, alkanolamine, and amino-acid families, together with the softwood lignin removal obtained after a sequential enzymatic–DES fractionation. Four descriptor-to-performance models were compared under identical leave-one-out, repeated k-fold, and grouped (leave-one-cluster-out and leave-one-family-out) cross-validation: a linear correlation, a 15-term second-order polynomial, a random forest (RF), and a Gaussian-process (GP) regressor with automatic relevance determination. All models fit the training set (in-sample R2 up to 0.89 for the RF) but generalize weakly: leave-one-out R2 is 0.10–0.13 for the GP and the RF, the polynomial is overparameterized and generalizes worse than the training mean, and no model predicts a chemical family that is absent from training better than the training mean. Gaussian-process relevance and ablation and random-forest permutation importance rank pHapp as the most informative descriptor, followed by α, whereas β and π* carry little consistent information, and the pKa of the hydrogen-bond donor carries the same information as pHapp; these results point to solvent acid strength, rather than hydrogen-bond donation alone, as the property most closely associated with lignin removal from enzymatically pretreated softwood. The GP provides calibrated predictive uncertainty: 96% of leave-one-out predictions and all four prospectively synthesized blind-test DESs fall within its 95% credible intervals. The results delimit what solvatochromic descriptors can and cannot predict for whole-biomass delignification and provide an uncertainty-aware screening tool and an open data set for DES design.

ACS Omega
National Laboratory of the Rockies (US), Western Michigan University (US), Tecnológico Nacional de México (MX), Technological Institute of Celaya (MX), Universidad de Burgos (ES)
U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy
Biofuel production and bioconversion
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon