Interpretable QSPR Modeling of Surfactant Characteristic Curvature
- 1Petroleum and Petrochemical College
- 2Chulalongkorn University
Abstract
Characteristic curvature (Cc) quantifies the intrinsic interfacial curvature tendency of surfactants under defined reference conditions and is a key input to the hydrophilic–lipophilic deviation (HLD) framework for predicting microemulsion phase behavior; however, experimental determination remains laborious, and interpretable molecular structure–Cc relationships across diverse surfactant chemistries are comparatively underdeveloped. Here, we developed a quantitative structure–property relationship (QSPR) framework using a curated literature-derived dataset of 63 anionic and nonionic surfactants, with molecular structures verified against the original sources and experimental values curated to improve consistency of the Cc endpoint, and benchmarked multiple tree-based regression algorithms using two-dimensional RDKit descriptors, fold-contained preprocessing, Yeo–Johnson target transformation, SHAP-based feature selection, and nested cross-validation. XGBoost achieved the strongest performance on 13-surfactants test set, with R 2 = 0.85, RMSE = 0.53, and MAE = 0.44 Cc units, with all metrics calculated on the inverse-transformed scale; the observed error was of the same order of magnitude as reported experimental uncertainty, supporting preliminary prediction within the represented chemical domain rather than replacing experimental Cc determination. SHAP and partial-dependence analyses identified electrotopological, topological, hydrophobicity-related, and other physicochemical descriptors associated with the model response, revealing nonlinear and threshold-like patterns that support chemically interpretable structure–property hypotheses without establishing causal mechanisms or experimentally validated design rules; the framework therefore offers a reproducible basis for prioritizing structurally related surfactants for subsequent experimental evaluation and provides a methodological reference for future Cc modeling with broader chemical coverage and prospective validation.
Related
- Interpretable Machine Learning-Based QSPR Modeling for Predicting Hydrogen Adsorption Performance of Ester Oil Impregnating Agents
- Predicting Glass Transition Temperature of Polyimide Homopolymers and Copolymers Using Interpretable QSPR-Based Machine Learning: The MFKit Framework
- Monte Carlo-based QSPR modeling of drop-weight impact sensitivity of energetic materials using hybrid SMILES–graph descriptors
- Surfactant-Toolkit and SurfBench-200: an open computation toolkit and a benchmark of tool-augmented language-model reasoning in surfactant science
- Salt Effects on the Growth of Mixed Anionic and Zwitterionic Surfactant Micelles
BibTeX
@misc{dulyadech2026interpretable,
title = {{Interpretable QSPR Modeling of Surfactant Characteristic Curvature}},
author = {Affan Dulyadech and Uthaiporn Suriyapraphadilok and Ampira Charoensaeng and Natthapong Sueviriyapan},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15009774/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009774/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Interpretable QSPR Modeling of Surfactant Characteristic Curvature
AU - Affan Dulyadech
AU - Uthaiporn Suriyapraphadilok
AU - Ampira Charoensaeng
AU - Natthapong Sueviriyapan
PY - 2026
DA - 2026/10/02
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009774/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009774/v1
AB - Characteristic curvature (Cc) quantifies the intrinsic interfacial curvature tendency of surfactants under defined reference conditions and is a key input to the hydrophilic–lipophilic deviation (HLD) framework for predicting microemulsion phase behavior; however, experimental determination remains laborious, and interpretable molecular structure–Cc relationships across diverse surfactant chemistries are comparatively underdeveloped. Here, we developed a quantitative structure–property relationship (QSPR) framework using a curated literature-derived dataset of 63 anionic and nonionic surfactants, with molecular structures verified against the original sources and experimental values curated to improve consistency of the Cc endpoint, and benchmarked multiple tree-based regression algorithms using two-dimensional RDKit descriptors, fold-contained preprocessing, Yeo–Johnson target transformation, SHAP-based feature selection, and nested cross-validation. XGBoost achieved the strongest performance on 13-surfactants test set, with R 2 = 0.85, RMSE = 0.53, and MAE = 0.44 Cc units, with all metrics calculated on the inverse-transformed scale; the observed error was of the same order of magnitude as reported experimental uncertainty, supporting preliminary prediction within the represented chemical domain rather than replacing experimental Cc determination. SHAP and partial-dependence analyses identified electrotopological, topological, hydrophobicity-related, and other physicochemical descriptors associated with the model response, revealing nonlinear and threshold-like patterns that support chemically interpretable structure–property hypotheses without establishing causal mechanisms or experimentally validated design rules; the framework therefore offers a reproducible basis for prioritizing structurally related surfactants for subsequent experimental evaluation and provides a methodological reference for future Cc modeling with broader chemical coverage and prospective validation.
ER -