(51g) Predicting Critical Micelle Concentrations for Surfactants Using Graph Convolutional Neural Networks

Surfactants are amphiphilic molecules that are widely used in consumer products [1], industrial processes [2], and biological research [3,4]. A critical property of a surfactant is the critical micelle concentration (CMC), at which surfactant molecules undergo cooperative self-assembly in solution [5]. The formation of micelles in surfactant solution can bring significant changes in properties such as surface tension, electrical conductivity, light scattering, and reaction mechanism [5,6]. Consequently, predicting CMCs is important for surfactant selection and design in applications such as drug delivery [7] and detergents [8].

Traditionally, CMCs are determined experimentally through techniques such as tensiometry, but this method is laborious and expensive [9]. As an alternative to experiments, computational methods such as molecular dynamics (MD) simulations [10, 11, 12] and quantitative structure-property relationships (QSPR) that are based on molecular descriptors [13, 14, 15] have been used to predict CMCs. These approaches have been shown to predict CMCs with relatively high accuracy, but they have several limitations. For instance, MD simulations usually require high computational cost and assumptions regarding the number of surfactants within a micelle [10, 11, 12], whereas QSPR models are often applicable to only a single class of surfactant and may need density functional theory (DFT) calculations to obtain quantum-chemical descriptors [16].

Recent advances in machine learning methods for molecular property prediction can help overcome some of these obstacles, specifically graph convolutional neural networks (GCNs) [17], since chemicals can be intuitively represented as molecular graphs that are natural inputs for GCNs. A GCN architecture can handle input molecules with different sizes without the need for artificial data manipulations and approximates the physical interactions between atoms through graph convolutions that aggregate features of adjacent atoms [17].

In this work, we present a simple GCN that can predict CMCs directly from the molecular structure of a surfactant monomer and test the ability of the model to generalize to a dataset containing nonionic, anionic, cationic, and zwitterionic surfactants [5]. We also perform saliency analysis [18] to interpret how atom types and surfactant substructures contribute to CMCs and compare these results with the physical rules that correlate structural information of surfactants to CMCs [6]. Following such rules, we propose a small set of new surfactants for which experimental CMCs are not available; for these molecules, we show that CMCs predicted with our GCN exhibited similar trends to those obtained from molecular simulations [19]. The findings of this research provide evidence that GCNs can be used for high-throughput screening of surfactants with desired self-assembly characteristics.

References:

[1] D. Myers, Surfactant Science and Technology: Third Edition, John Wiley and Sons, 2005.
[2] A. Barati, A. Najafi, A. Daryasafar, P. Nadali, H. Moslehi, Adsorption of a new nonionic surfactant on carbonate minerals in enhanced oil recovery: Experimental and modeling study, Chem. Eng. Res. Des. 105 (2016) 55–63.
[3] G.P. Kumar, P. Rajeshwarrao, Nonionic surfactant vesicular systems for effective drug delivery—an overview, Acta Pharm. Sin. B. 1 (2011) 208–219.
[4] F. Gallou, N.A. Isley, A. Ganic, U. Onken, M. Parmentier, Surfactant technology applied toward an active pharmaceutical ingredient: more than a simple green chemistry advance, Green Chem. 18 (2015) 14–19.
[5] M.J. Rosen, J.T. Kunjappu, Surfactants and Interfacial Phenomena: Fourth Edition, John Wiley and Sons, 2012.
[6] J. Israelachvili, Intermolecular and Surface Forces, Elsevier Inc., 2011.
[7] V.P. Torchilin, Structure and design of polymeric surfactant-based drug delivery systems, J. Control. Release. 73 (2001) 137–172.
[8] K.C. Cheng, Z.S. Khoo, N.W. Lo, W.J. Tan, N.G. Chemmangattuvalappil, Design and performance optimisation of detergent product containing binary mixture of anionic-nonionic surfactants, Heliyon. 6 (2020) e03861.
[9] N. Scholz, T. Behnke, U. Resch-Genger, Determination of the Critical Micelle Concentration of Neutral and Ionic Surfactants with Fluorometry, Conductometry, and Surface Tension—A Method Comparison, J. Fluoresc. 28 (2018) 465–476.
[10] C.G. Gahan, S.J. Patel, M.E. Boursier, K.E. Nyffeler, J. Jennings, N.L. Abbott, H.E. Blackwell, R.C. Van Lehn, D.M. Lynn, Bacterial Quorum Sensing Signals Self-Assemble in Aqueous Media to Form Micelles and Vesicles: An Integrated Experimental and Molecular Dynamics Study, J. Phys. Chem. 2020 (2020) 3628.
[11] A. Vishnyakov, M.T. Lee, A. V. Neimark, Prediction of the critical micelle concentration of nonionic surfactants by dissipative particle dynamics simulations, J. Phys. Chem. Lett. 4 (2013) 797–802.
[12] A.P. Santos, A.Z. Panagiotopoulos, Determination of the critical micelle concentration in simulations of surfactant systems, J. Chem. Phys. 144 (2016) 044709.
[13] K. Roy, H. Kabir, QSPR with extended topochemical atom (ETA) indices: Modeling of critical micelle concentration of non-ionic surfactants, Chem. Eng. Sci. (2012).
[14] T. Gaudin, P. Rotureau, I. Pezron, G. Fayet, New QSPR models to predict the critical micelle concentration of sugar-based surfactants, Ind. Eng. Chem. Res. (2016).
[15] X. Li, G. Zhang, J. Dong, X. Zhou, X. Yan, M. Luo, Estimation of critical micelle concentration of anionic surfactants with QSPR approach, J. Mol. Struct. THEOCHEM. 710 (2004) 119–126.
[16] P.D.T. Huibers, V.S. Lobanov, A.R. Katritzky, D.O. Shah, M. Karelson, Prediction of critical micelle concentration using a quantitative structure-property relationship approach. 2. Anionic surfactants, J. Colloid Interface Sci. 187 (1997) 113–120.
[17] Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, P.S. Yu, A Comprehensive Survey on Graph Neural Networks, ArXiv. (2019).
[18] J. Adebayo, J. Gilmer, M. Muelly, I. Goodfellow, M. Hardt, B. Kim, Sanity Checks for Saliency Maps, Adv. Neural Inf. Process. Syst. 2018-December (2018) 9505–9515.
[19] T. Jin, S.J. Patel, R.C. Van Lehn, Molecular simulations of lipid membrane partitioning and translocation by bacterial quorum sensing modulators, PLoS One. 16 (2021) e0246187.