Automated inference of disease mechanisms in patient-hiPSC-derived neuronal networks

Abstract

Human induced pluripotent stem cells (hiPSCs)-derived neurons offer a valuable platform for studying neurological disorders in a patient-specific manner. These neurons can be rapidly differentiated into excitatory neuronal networks, whose activity is measurable using multi-electrode arrays (MEAs). Neuronal networks derived from patients exhibit distinct characteristics, reflecting underlying pathological molecular mechanisms. However, elucidating these mechanisms traditionally requires extensive and hypothesis-driven additional experiments. Computational models can link observable network activity to underlying molecular mechanisms by identifying biophysical model parameters that simulate the activity, but this identification process is challenging. Here, we address this challenge by using simulation-based inference (SBI), a machine-learning approach, to automatically identify the full range of model parameters able to explain the patient-derived MEA activity. Our study demonstrates that SBI can accurately identify ground-truth parameters in simulated data, and successfully estimate the parameters that replicate the network activity of healthy hiPSC-derived neuronal networks. Furthermore, we show that SBI can pinpoint molecular mechanisms affected by pharmacological agents and identify key disease mechanisms in neuronal networks derived from patients. These findings underscore the potential of SBI to automate and enhance the identification of disease mechanisms from MEA measurements, offering a robust and scalable method for advancing research with hiPSC-derived neuronal networks.