Identifying the Control Knobs of Cytoskeletal Flow using Simulations and Representation Learning

Dynamic contractile networks of actomyosin actin filaments, myosin motors, and cross-linking proteins can drive large-scale flow that polarizes cells during cytokinesis in C. elegans embryos. This flow is mainly driven by the myosin motor gradient. However, it is challenging to elucidate the role of myosin in controlling the magnitude of this macroscopic flow. In this work, we combine physical modeling and representation learning (a convolutional autoencoder) to identify a useful latent representation that detects the mechanisms of contractility contributing to large-scale flow. This latent representation demonstrates the importance of coupling between myosin and the buckling of actin filaments. The methodologies developed in this work can be generalized and combined with either experimental imaging or computer simulations to investigate the physics of dynamic contractile networks.