Machine Learning Analysis for Identification of Cell Shape Metrics Associated with Stem Cell Differentiation in Nanofiber Scaffolds

Abstract

Cell shape has been demonstrated to be closely related to cell function and may be an important predictor of cell fate. Many metrics are available to describe cell shape, however the relationship of these metrics to cell fate are not well understood. Specifically, nanofiber scaffold structures have been demonstrated to uniquely induce osteogenic differentiation of human bone marrow stromal cells (hBMSCs) and alter cell shape, similarly to chemically induced differentiation. This phenomenon occurs after only 1 day of culture, and may allow for early prediction of stem cell fate. In this study, we aim to classify cells on nanofiber scaffolds versus flat film substrates based on their shape metrics and correlate these metrics to functional outcomes of the cells. We have developed computational tools based on support vector machine to identify cell morphological features associated with nanofiber induced differentiation of hBMSCs. To accurately classify cells based on shape and to determine the most significant cell shape metrics we have utilized the supercell method to account for cell shape heterogeneity as well as a jackknife method to test the robustness of our classifications.