Maybe trying to reverse engineer it using GNNs until they produce the same output(within some error margin) for various arbitrary inputs? Dunno.
However GNNs don’t really capture the underlying mechanisms very well. Our brains use different kinds of Neurons which also use many different neural transmitters, their location in 3D space matters, as do stuff like blood alcohol content, O2 levels, adrenaline, signal propagation delays, etc it’s a stupidly complicated system even above and beyond the number of neurons.