Single-cell computational modeling and AI

Computational modeling of individual neurons opens up the possibility for researchers to study neurons with a level of specificity not otherwise available. Dr Hue Ye’s presentation about restoring conductance to a locally demyelinated axon showed how when there are physical limitations to an experiment, computational models can be used to gather information instead. Applying an electromagnetic signal would disrupt accurate recording at the nodes of Ranvier, so the neuron was modeled in the NEURON simulation environment.

The ability to model neural systems on a computer has been a point of discussion for neuroscientists for a long time. A recent article from TIME explores the necessity of a “virtual cell” and its potential applications (https://time.com/7324119/what-is-virtual-cell/). Some scientists think it could be possible to make a virtual cell so advanced that individual proteins and functional disruptions are able to be visualized and observed. The article also addresses the potential for AI to help with developing a virtual cell. Recent approaches have turned away from modeling equations by hand, and towards feeding information to specialized AI programs instead. These programs learn quickly from the cell data, and are eventually able to predict the behavior of cells belonging to species they had not yet encountered.

The intersection of neuroscience and machine learning is fascinating to me, because especially in regard to this example, it seems like we’ve come full circle. The structure and function of the brain informs our research on machine learning and neural networks, and the AI programs that have resulted from that information are now assisting with neuroscience research. The article suggests that this AI driven virtual cell concept could be used for drug development, but as it gets more advanced I think its applications will expand to include a variety of topics like pathologies and electrophysiology.