AI and Inverse Methods for Building Digital Twins in Neuroscience
Big data in neuroscience is driving the development of methods seeking to predict the state of neurons and neuronal circuitry. Model-based methods (e.g., data assimilation) and data-based methods (e.g. reservoir computing, AI) have had great success at constructing digital twins of neurons and small circuits such as central pattern generators. These digital twins have begun making remarkably accurate predictions of observable quantities such as membrane voltage oscillations and are unique in predicting the dynamics of variables inaccessible to experiment (ionic current waveforms, gate probabilities). Successful digital twins have the potential to advance fundamental biology by revealing channelopathies, they can help clinicians make diagnosis, provide therapies for chronic diseases when embodied in-silico as neurostimulation devices (e.g., neuronal pacemaker) or brain-machine interfaces. The topic is aimed at providing a forum where various techniques for constructing digital twins in