276 — Estimating the impact of structural directionality: How reliable are undirected connectomes?

Kale, Zalesky, & Gollo (1801.01651)

Read on 23 May 2018
#neuroscience  #graphs  #networks  #directed-graph  #connectome  #c-elegans  #noninvasive  #MRI  #fMRI  #directionality 

Many of the methods that we commonly use to derive connectome graphs can determine the direction of data flow: For example, in electron microscopy, the synapses and the vesicles themselves are visible, and we can identify pre- and post-synaptic partners.

But the problem with EM is that you have to cut up the brain really really small. And people get really upset when you do that to their brains.

And so when we create connectomes in a noninvasive way, we use methods like MRI, where there is absolutely no way we could figure out in which direction individual synapses point.

This research looks at directed connectomes — namely, from studies in some mammals as well as the full connectome of C. elegans — and, by stripping away this directionality, aims to find out what we can still learn about a connectome when we do not have direction information.

In cases where directionality was estimated (e.g. fMRI), higher-area-count parcellations tended to have a higher proportion of unidirectional edges (which makes sense). As a result of this, these higher-acuity brain graphs were much more susceptible to large changes in graph invariants when directed edges were replaced with undirected edges.

In the case of the C. elegans nematode, these changes still had a large, noticeable impact on the recorded global graph measures (duh) but because of the nature of the worm’s connectome and the fact that it is represented in highly structured and highly confident data (i.e. we do know the exact connections in this graph, rather than the MRI approximations) these changes made less of an overall impact.