The recent news of the success of Google’s AI platform-AlphaFold2- in predicting protein structure with unprecedented accuracy has electrified the field of biology and renewed interest in the application of deep-learning in discovery science. By predicting, to a great accuracy, the relationship of amino acid sequences and the 3D geometry of the resulting protein, google has made way for the manipulation of the bio-molecular world at an unprecedented scale.

Alongside, the demonstration has drawn attention to a broad class of biological problems- the structure-function mapping. Biological systems execute their myriad functions by possessing structures that can manipulate the environment in specific ways. Neuroscience, for example, has its own version of the structure-function problem. In the case of the brain, the structure corresponds to the ordered, hierarchical organization of neurons and the function relates to the correlations of activity patterns, most typically measured as BOLD activations.

As with proteins, the structure-function problem in neuroscience is complicated by the mind-bogglingly non-linear and networked nature of interaction between neurons. This fact can be gauged from the observation that SC explains, at best, only about half of the variance in FC. This discrepancy is further widened in case of functional connections between areas that lack any direct structural connectivity.

Ofcourse, understanding the link between SC and FC at the individual level has important consequences for the treatment of disease, many of which manifest as impairments in functional connections between brain areas.

So can we have a neuroscience equivalent of Alpha Fold ?

References:

Suárez, L. E., Markello, R. D., Betzel, R. F., & Misic, B. (2020). Linking structure and function in macroscale brain networks. Trends in Cognitive Sciences.