Scalable Extraction of Information from Spatio-Temporal Patterns of Chemoresponsive Liquid Crystals Using Topological Data Analysis

Abstract

Chemoresponsive liquid crystals (LCs) can be engineered to generate information-rich optical responses upon exposure to gas contaminants. We investigate the use of topological descriptors to extract information from these complex responses and with this facilitate sensor design and understand physical phenomena that govern responses. We provide a holistic view of topological descriptors using Minkowski functionals and fractal analysis and show how descriptors can be used for unsupervised (clustering, visualization) and supervised (regression, classification) machine learning (ML) tasks. Specifically, by using high-throughput, experimental data for LC films exposed to diverse contaminants, we show that topological descriptors can be used to effectively detect outliers and predict contaminant concentrations using simple ML models. Notably, these models achieve comparable accuracies to those of powerful convolutional neural networks, but with much lower computational times (from hours to seconds) and using less sophisticated computing hardware. This scalability enables high-resolution, space-time data analysis.