An important feature of complex networks that can help to understand their internal organization is community structure. In particular, to recognize such structures in brain networks could be crucial, as the brain functioning is thought to be based on modular organization. Moreover, brain networks are intrinsically multilayer, which essentially means they can vary in time, in frequency or in other domain depending on the topology of the network. In the last decades, some multilayer clustering algorithm has been developed with the aim to identify communities in networks of different nature. However, there is still no agreement about which one is the most reliable, and a way to test and compare these algorithms under a variety of conditions is lacking. With this project, we aim to perform a comparative analysis between different multislice clustering algorithms, evaluating their performances by means of ad-hoc implemented benchmark graphs characterized by properties that will cover a wide range of conditions. Results will seek to provide some guidelines about the choice of the more appropriate algorithm according to the different conditions. As a proof of concept, the algorithms under exam will be also applied to brain functional connectivity networks obtained from EEG signals collected during a working memory task, and we will hopefully prove that the clustering of scalp electrodes agrees with the results of the simulation study.
This project aims to provide some guidelines about the use of clustering algorithms on multislice brain networks. A novel approach will be pursued since the algorithms will be tested on innovative multilayer graphs models with flexible features. This flexibility, together with the possibility of simulating stable and dynamic communities, will guarantee a deep analysis of the behaviour of the clustering algorithms that is actually lacking.
An increasing number of studies pointed out how much important is exploiting and studying multilayers networks to understand human brain functioning and organization, and models able to consider different characteristics of these networks are needed. The model that will be developed thanks to this project will constitute a first dowel, and will surely benefit of further studies that can make it more and more similar to real brain networks (for example introducing hierarchical structure, or constraints on degrees of the nodes). The final aim would be finding important relations among structures, dynamics, cognition and health of the human brain to gain a more thorough knowledge in this field.
The work will be oriented to application on EEG based brain networks, so that simulated networks will have their typical features. However further studies might use the tool that will be developed to test the algorithms in conditions belonging to other fields. Thus, for example if one is interested in exploring community structure in social networks, the same generator can be used setting a higher number of nodes and clusters. Furthermore, if a new clustering algorithm will be proposed, it could be tested directly on these benchmark graphs and it could be put in relation with the results that will be issued by this study. In this way, guidelines about the choice of clustering algorithms will be continuously updated, and every user will be able to choose the most performing one depending on the field of application.