Mathematical models are powerful means to encode prior knowledge. While valuable in data-sparse settings, this limits unbiased discovery. We apply data-driven methods to enable the unbiased inference of explainable dynamic models from large perturbational screens.