Cell membrane mechanics using a simple to fabricate microwell chip and deep learning-assisted automated AFM analysis

Atomic Force Microscopy (AFM) being inherently slow and analysis heavy, becomes challenging for scaling up. Addressing this, we take a two-fold approach; first we introduce an easy-to-fabricate reusable poly(dimethylsiloxane)-based array that consists of micron-sized traps for single-cell trapping and second, we apply a deep-learning method directly on the extracted curves to facilitate and automate the analysis. Our approach is validated using suspended cells which often require specific holders or adhesive molecules due to their tendency to slip from the surface. Using nanoindentation, cell cortex stiffness alterations, under the influence of three different drugs that inhibit myosin activity, are revealed. We then apply machine learning models to extract membrane stiffness directly from the raw data as well for binary (presence/absence of drugs) and multiclass classification (different drug types). The proposed analysis resulted in a Coefficient of Determination of 0.47 for the regression problem while for the binary and multiclass classification the analysis resulted in an Area Under the Curve score of 0.91 and accuracy scores exceeding 0.9 respectively, for each individual drug class. Overall, the versatility to fabricate the microwells in conjunction with the automated analysis and classification could find wide-range applications spanning from to basic cell-based assays to drug screening.