Develop simulation-supervised machine learning for biological imaging by combining differentiable programming, biophysical simulation, and domain-adaptation to train models on realistic synthetic data that generalize to unlabeled experimental datasets.