Researchers develop a more precise method for identifying equivalent neurons

Comparing brains to understand how neurons connect and how these connections can vary between species or individuals is one of the great areas of research in neuroscience. Now, a research team from the Department of Chemical Engineering at the Universitat Rovira i Virgili (URV) has developed a probabilistic model that allows different complex networks to be aligned at the same time in order to find common patterns. This method, which has been published in the scientific journal Nature Communications, not only substantially improves on current methods, but also represents a revolutionary approach to the problem that will allow us in the near future to align networks of hundreds of thousands of neurons in an effective way, something that cannot be done at present.

This method is based on a simple but biologically plausible idea: it assumes that all the networks to be aligned are error-laden copies of the same underlying structure, like a kind of blueprint. It then reconstructs this common pattern from the available observations. This allows researchers not only to align the networks, but also to estimate the probability that each node (which could be a neuron or a user) in one network corresponds to a given node in the others.

"Our main motivation was to be able to compare connectomes, i.e. the maps of neuronal connections within the brain. But to do so, we first needed to know which neurons are equivalent in each brain," explains Marta Sales-Pardo, a researcher at the Department of Chemical Engineering involved in the research. This task is particularly complex when dealing with brains from different people or even different species and when there is no clear contextual information about each neuron.

The method has been validated with real data from connectomes from the C. elegans nematode at different stages of development, from the larva of the Drosophila melanogaster (known as the vinegar fly) and also with email networks. In all three cases, their technique has demonstrated a precision far superior to existing methods. "We've been able to align up to five networks at a time and we've done it better than any other technique hitherto developed," says Roger Guimerà, ICREA professor in the same department.

In addition to increasing precision, the probabilistic approach offers a fundamental advantage: it makes explicit hypotheses about how the data are generated. This makes it possible to adapt the model to different types of networks, to add contextual information such as the categories of nodes (for example, the type of neuron) and to better interpret the results. "With this method, we not only look for the best possible alignment, but we also estimate the probability that each correspondence is correct, which is key to making reliable inferences," they add.

Although the work is set in the context of neuroscience, the method is applicable to any network where there may be equivalence patterns. The team has also tested it on email networks: "We started by knowing who each user was so we could determine that the method correctly found who was who through the connections," explains Teresa Lázaro, who is also a researcher at the URV's Department of Chemical Engineering. The research could also be applied to networks of interactions between proteins and thus be useful for identifying unknown functions in new or pathogenic organisms, or for detecting suspicious patterns in financial or security networks.

The results of the research are promising, as they "allow us to compare networks in a microscopic way, far beyond the general indicators that have been used until now", concluded the researchers.