AIggregate: A Machine Learning Approach for Classifying Micelle Shape

Abstract

In this work we develop AIggregate, a machine learning tool for classifying the shapes of micelles, which are clusters of surfactant molecules self assembled in aqueous solutions due to their unique amphiphilic character. For the majority of these systems, as the concentration of the surfactant increases, the micellar shape changes from spherical to elongated at a specific value defining the second critical micellar concentration (CMC). Known methods aiming to classify the micelles’ shape and to specify the second CMC with molecular modeling use heuristic approaches. These are based on shape parameters like asphericity, acylindricity, and anisotropicity. We expand upon this approach by applying machine learning and deep learning architectures to classify the shape of molecular assemblies. To achieve our goal, AIggregate uses both a point cloud representation of the micelle, where each atom or group of atoms constitutes a point in the cloud, as well as shape parameters of the assembly. We found that these methods significantly improve classification accuracy over the heuristic approach, with the deep learning, point-cloud-based method offering the maximum efficiency among the examined methods.