Genome-wide profiling of the hierarchical control of autophagy dynamics using deep learning
Recycling of cellular components through autophagy maintains homeostasis in dynamic nutrient environments, and its dysregulation is linked to several human disorders. Although extensive research has characterized the core mechanisms of autophagy, limited insight into its systems-wide dynamic control has hampered predictive modeling and effective in vivo manipulation. In this study, we mapped the genetic network that controls both the dynamic activation and inactivation of autophagy during nitrogen changes, using a combination of time-resolved high-content imaging, deep learning, and latent feature analysis. This approach generated a comprehensive genome-wide profiling repository, termed AutoDRY, categorizing 5919 mutants based on their nutrient response kinetics and differential contributions to autophagosome formation and clearance. Integrating these profiles with functional and genetic network data unveiled a hierarchical and multi-layered control of autophagy, identifying new regulatory aspects of the core machinery and established nutrient-sensing pathways. By leveraging multi-omics resources and explainable machine learning to predict genetic perturbation effects and infer new regulatory mechanisms, we identified the retrograde pathway as a pivotal, time-varying autophagy modulator through transcriptional tuning of core genes. By charting the systems-wide dynamical control of autophagy, we have laid the groundwork for connecting the complexity of genome-wide influences with specific core mechanisms. This represents a significant advancement in studying complex genetic phenotypes, guides functional genomics of dynamic cellular processes in any organism, and provides a powerful starting point for hypothesis-based research on autophagy.