Prediction of Golgi Polarity in Collectively Migrating Epithelial Cells Using Graph Neural Network.

In the stationary epithelium, the Golgi apparatus assumes an apical position, above the cell nucleus. However, during wound healing and morphogenesis, as the epithelial cells start migrating, it relocalizes closer to the basal plane. On this plane, the position of Golgi with respect to the cell nucleus defines the organizational polarity of a migrating epithelial cell, which is crucial for an efficient collective migration. Yet, factors influencing the Golgi polarity remain elusive. Here, we constructed a graph neural network-based deep learning model to systematically analyze the dependency of Golgi polarity on multiple geometric and physical factors. In spite of the complexity of a migrating epithelial monolayer, our simple model was able to predict the Golgi polarity with 75% accuracy. Moreover, the model predicted that Golgi polarity predominantly correlates with the orientation of maximum principal stress. Finally, we found that this correlation operates locally since progressive coarsening of the stress field over multiple cell-lengths reduced the stress polarity-Golgi polarity correlation as well as the predictive accuracy of the neural network model. Taken together, our results demonstrate that graph neural networks could be a powerful tool toward understanding how different physical factors influence collective cell migration. They also highlight a previously unknown role of physical cues in defining the intracellular organization.