Predicting Critical Micelle Concentrations for Surfactants Using Graph Convolutional Neural Networks.

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Engineering

Background:

• Surfactants are essential amphiphilic molecules with applications spanning consumer products to industrial processes.
• Determining the critical micelle concentration (CMC) is crucial for understanding surfactant self-assembly, but experimental methods like tensiometry are time-consuming and costly.

Purpose of the Study:

• To develop a novel computational method for predicting surfactant CMCs directly from molecular structure.
• To establish a machine learning model capable of accurate and generalizable CMC prediction across various surfactant types.

Main Methods:

• Utilized graph convolutional neural networks (GCNs) to encode surfactant molecular structures as graphs.
• Trained the GCN model on a comprehensive dataset of experimental CMC values.
• Employed molecular saliency maps to interpret model predictions and identify key structural contributors to CMC.

Main Results:

• The GCN model achieved high accuracy in predicting CMCs, outperforming previous methods on an inclusive dataset.
• The model demonstrated generalization capabilities across anionic, cationic, zwitterionic, and nonionic surfactants.
• Molecular saliency maps revealed structure-CMC relationships consistent with established physical principles.

Conclusions:

• GCNs provide an accurate and efficient tool for predicting surfactant CMCs from molecular structure.
• This computational approach facilitates high-throughput screening of surfactants for desired self-assembly properties.
• The findings pave the way for accelerated discovery and design of novel surfactants.