Pattern recognition of drying lysozyme–glucose droplets using machine learning classifiers

Out-of-equilibrium processes, such as sessile droplet drying, often result in distinctive macroscopic residual patterns in systems containing molecules, proteins, and colloids. Protein-glucose mixtures are particularly effective models for studying the behavior of complex fluids containing biomolecules. This study investigates the drying patterns of lysozyme droplets with varying initial glucose concentrations. Without glucose, the crack patterns are chaotic and dispersed throughout the droplet. Interestingly, cracks predominantly form around the droplet edges at intermediate glucose concentrations, while the deposits become uniform and crack-free at high glucose concentrations. To understand and classify the unique patterns related to the initial compositional changes, we developed an automated pattern recognition pipeline. We used two methods for analyzing images captured throughout the drying process. The first method involved extracting statistical textural parameters from the images as quantitative features for machine learning classifiers. The second method utilized a neural network-based classifier to directly classify the images, achieving an accuracy of 97%. The results demonstrate the effectiveness of using images from the entire drying process, not just the final images, for pattern classification. This approach may be useful in gaining a fundamental understanding of unique crack pattern that emerge when glucose is added to a protein solution. • Sessile drying reveals unique macroscopic patterns in lysozyme-glucose droplets. • Pattern is chaotic without glucose, while high glucose leads to crack-free deposit. • At intermediate glucose levels, cracks only appear in the periphery of the droplets. • An automated pattern recognition is developed to classify the drying patterns. • It achieves an accuracy of 93%–97% in correlating patterns and initial compositions.