Characterising fate transitions in cell cultures using data-driven force inference.

The molecular mechanisms driving the dynamics and patterning of cell cultures and tissues are very complex, especially during cell fate transitions. However, there is a wealth of microscopy data which can be harnessed to help us understand some of the hidden driving forces. In this work we are interested in the epithelial-to-mesenchymal transition which underlies key biological processes such as early embryonic development as well as tumorigenesis. Taking a top-down approach, we infer the forces governing cell dynamics by partitioning them into constitutive parts: pairwise forces, chemical gradient forces and intrinsic noise, among others. We approximate each force term using a neural network and train a model to predict the motion of cells, tracked using nuclear markers. Using this data-driven approach we can point toward which terms in the equations of motion of cells change when a cell changes state, which we use to simulate the behaviour of a collection of such cells, grown in different densities and chemical environments. We believe this framework will be useful in understanding the mechanisms driving complex pattern formation in tissues and tumours, which arise from the intrinsic differences in the forces governing different cell types.