Artificial Neural Network-Aided Computational Approach for Mechanophenotyping of Biological Cells Using Atomic Force Microscopy

Abstract

The artificial neural network (ANN) based models have shown the potential to provide alternate data-driven solutions in disease diagnostics, cell sorting and overcoming AFM-related limitations. Hertzian model-based prediction of mechanical properties of biological cells, although most widely used, has shown to have limited potential in determining constitutive parameters of cells of uneven shape and non-linear nature of force-indentation curves in AFM-based cell nanoindentation. We report a new artificial neural network-aided approach, which takes into account, the variation in cell shapes and their effect on the predictions in cell mechanophenotyping. We have developed an artificial neural network (ANN) model which could predict the mechanical properties of biological cells by utilizing the force vs. indentation curve of AFM. For cells with 1 µm contact length (Platelets), we obtained a recall of 0.97 ± 0.03 and 0.99 ± 0.0 for cells with hyperelastic and linear elastic constitutive properties respectively with a prediction error of less than 10%. Also, for cells with 6-8 µm contact length (RBCs), we obtained the recall of 0.975 in predicting mechanical properties with less than 15% error. We envisage that the developed technique can be used for better estimation of cells' constitutive parameters by incorporating cell topography into account.