Network-aware self-supervised learning enables high-content phenotypic screening for genetic modifiers of neuronal activity dynamics

Abstract

High-throughput phenotypic screening has historically relied on manually selected features, limiting our ability to capture complex cellular processes, particularly neuronal activity dynamics. While recent advances in self-supervised learning have revolutionized the ability to study cellular morphology and transcriptomics, dynamic cellular processes have remained challenging to phenotypically profile. To address this limitation, we developed Plexus, a self-supervised model specifically designed to capture and quantify network-level neuronal activity dynamics. Unlike existing phenotyping tools that focus on static readouts, Plexus leverages a network-level cell encoding method, which enables it to efficiently encode dynamic neuronal activity data into rich representational embeddings. In turn, Plexus achieves state of the art performance in detecting phenotypic changes in neuronal activity. We validated Plexus using a comprehensive GCaMP6m simulation framework and demonstrated its enhanced ability to classify distinct neuronal activity phenotypes compared to traditional signal-processing approaches. To enable practical application, we integrated Plexus with a scalable experimental system utilizing human iPSC-derived neurons equipped with the GCaMP6m calcium indicator and CRISPR interference machinery. This integrated platform successfully identified nearly seventeen times as many distinct phenotypic changes in response to genetic perturbations compared to conventional signal processing methods, as demonstrated in a 52-gene CRISPRi screen across multiple iPSC lines. Using this framework, we identified potential genetic modifiers of aberrant neuronal activity in frontotemporal dementia, illustrating its utility for understanding complex neurological disorders.