Abstract
The brain is an extraordinarily complex system that facilitates the optimal integration of information from different regions to execute its functions. With the recent advances in technology, researchers can now collect enormous amounts of data from the brain using neuroimaging at different scales and from numerous modalities. With that comes the need for sophisticated tools for analysis. The field of network neuroscience has been trying to tackle these challenges, and graph theory has been one of its essential branches through the investigation of brain networks. Recently, topological data analysis has gained more attention as an alternative framework by providing a set of metrics that go beyond pairwise connections and offer improved robustness against noise. In this handson tutorial, our goal is to provide the computational tools to explore neuroimaging data using these frameworks and to facilitate their accessibility, data visualisation, and comprehension for newcomers to the field. We will start by giving a concise (and by no means complete) overview of the field to introduce the two frameworks and then explain how to compute both wellestablished and newer metrics on restingstate functional magnetic resonance imaging. We use an opensource language (Python) and provide an accompanying publicly available Jupyter Notebook that uses the 1000 Functional Connectomes Project dataset. Moreover, we would like to highlight one part of our notebook dedicated to the realistic visualisation of high order interactions in brain networks. This pipeline provides threedimensional (3D) plots of pairwise and higherorder interactions projected in a brain atlas, a new feature tailormade for network neuroscience.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
Neuroscience is still a young research field, with its emergence as a formal discipline happening only around 70 years ago (Cowan et al. 2000). The field has since mushroomed, and much of our current knowledge about the human brain’s neurobiology was made possible by the rapid advances in technologies to investigate the brain in vivo at highresolution and different scales. An example is magnetic resonance imaging (MRI), which allows us to measure regional characteristics of the brain’s structure noninvasively and may also be used to assess anatomical and functional interactions between brain regions (Rosen and Savoy 2012; Sizemore et al. 2018). This expansion in the field led to an exponential increase in data size and complexity. To analyse and interpret this ‘big data’, researchers had to develop robust theoretical frameworks. Complex network science was brought to neuroscience and has been increasingly used to study the brain’s intricate communication and wiring (Bassett and Sporns 2017; Sporns 2018). The resulting field—network neuroscience—aims to see the brain through an integrative lens by mapping and modelling its elements and interactions (Bassett and Sporns 2017; Fornito et al. 2016).
One of the main theoretical frameworks from complex network science used to model, estimate, and simulate brain networks is graph theory (Gross and Yellen 2003; Bullmore and Sporns 2009). A graph is comprised of a set of interconnected elements, also known as vertices and edges. Vertices (also known as nodes) in a network can, for example, be brain areas, while edges (also known as links) are a representation of the functional connectivity between pairs of vertices (Sporns 2018). Various imaging modalities are available to reconstruct the brain network (Hart et al. 2016; Bullmore and Sporns 2009). The focus of this handson paper will be restingstate functional MRI (rsfMRI). As the name suggests, rsfMRI indirectly measures brain activity while a subject is at rest (i.e., does not perform any task). This type of data provides information about spontaneous brain functional connectivity (Raichle 2011). Functional connectivity is often operationalised by a statistical dependency (usually a Pearson correlation coefficient) between signals measured from anatomically separated brain areas (Rosen and Savoy 2012; Smith et al. 2013). An indepth explanation of rsfMRI and functional connectivity is out of the scope of our manuscript. However, considering the focus on this type of data here, we recommend readers who are not familiar with this imaging method to read Lee et al. (2013); van den Heuvel and Hulshoff Pol (2010); Smith et al. (2013); Smitha et al. (2017) for a comprehensive overview.
Several descriptive graph metrics^{Footnote 1} (Do Carmo 2016) can be calculated to describe the brain network’s characteristic; examples include the degree or the total number of connections of a vertex and the path length (number of intermediate edges) between two vertices (Fornito et al. 2016; Hallquist and Hillary 2018). These metrics have consistently allowed researchers to identify nonrandom features of brain networks. A key example is the groundbreaking discovery that the brain (like most other realworld networks) follows a ‘smallworld network’ architecture (Bassett and Bullmore 2017; Bassett and Sporns 2017; Watts and Strogatz 1998). This refers to the phenomenon that, to minimise wiring cost while simultaneously maintaining optimal efficiency and robustness against perturbation, the brain network obeys a balance between the ability to perform local processing (i.e., segregation) and combining information streams on a global level (i.e., integration).
Network neuroscience has thereby offered a comprehensive set of analytical tools to study not only the local properties of brain areas but also their significance for the entire brain network functioning. Using graph theory, many insights have been gathered on the healthy and diseased brain neurobiology (Farahani et al. 2019; Hallquist and Hillary 2018; Hart et al. 2016; Sporns 2018).
Another perspective on the characteristics of the brain network can be provided by (algebraic) topological data analysis (TDA), by analysing the interactions between a set of vertices beyond the ‘simple’ pairwise connections (i.e., higherorder interactions). With TDA, one can identify a network’s ‘shape’ and its invariant properties [i.e., coordinate and deformation invariances (Zomorodian 2005; Offroy and Duponchel 2016)]. Thus, as we will illustrate along with the manuscript, TDA often provides more robustness against noise than graph theoretical analysis (Blevins and Bassett 2020; Blevins et al. 2021), which can be a significant issue in imaging data (Sizemore et al. 2019; Liu 2016; Greve et al. 2013). Although TDA has only recently been adopted in network neuroscience (Curto and Itskov 2008; Singh et al. 2008), it has already shown exciting results on rsfMRI (Expert et al. 2019; Curto 2017). For example, grouplevel differences in network topology have been identified between healthy subjects that ingested psilocybin (psychedelic substance) and the placebo group (Petri et al. 2014) and between attentiondeficit/hyperactivity disorder children and typically developing controls (GraciaTabuenca et al. 2020). A limitation of this framework is that the complexity and level of mathematical abstraction necessary to apply TDA and interpret the results might keep clinical neuroscientists without prior mathematical training from using it. Moreover, the highorder interaction structure that emerges from TDA analysis is often challenging to visualise realistically and understandably. Despite technical constraints, TDA allows us to deal with high order and large combinatorial coding capacity properly.
Therefore, we would like to facilitate the use of network neuroscience and its constituents graph theory and TDA by the general neuroscientific community by providing a stepbystep tutorial on how to compute different metrics commonly used to study brain networks and realistic highorder network plots. We offer a theoretical and experimental background of these metrics and include code blocks in each section to explain how to compute the different metrics. We also list several additional resources (Tables 1 and 2) of personal preference (and by no means complete), including a Jupyter Notebook that we created to accompany this handson tutorial publicly available on GitHub and Zenodo (Centeno and Santos 2021) (see Table 1, under the Jupyter Notebooks section—Notebook for network and topological analysis in neuroscience).
Our work differs from previous literature (Hallquist and Hillary 2018; Otter et al. 2017) since we describe the concepts central to graph theory and TDA and provide an easytograsp stepbystep tutorial on how to compute these metrics using an easily accessible, opensource computer language. Furthermore, we offer new 3D visualisations of simplicial complexes and TDA metrics in the brain that may facilitate the application and interpretation of these tools. Finally, we would like to stress that even though this tutorial focuses on rsfMRI, the main concepts and tools discussed in this paper can be extrapolated to other imaging modalities, biological or complex networks.
Since graph theory has been extensively translated for neuroscientists elsewhere, we refer the reader to the book in Fornito et al. (2016). This tutorial mainly focused on the topics covered in chapters 3, 4, 5, and the particular sections of chapters 6, 7, 8, and 9 about assortativity, shortest paths and the characteristic path length, the clustering coefficient, and modularity. In the second part of the tutorial, we explore handson TDA metrics, providing a summary of both theoretical and neuroscientific aspects with the calculations used in our work. We believe that our tutorial, which is far from being exhaustive, can make this emerging branch of network and topological neuroscience accessible to the reader. The codes we provide only require the knowlege of the functional connectivity matrix. For our realistic 3D visualisation of simplicial complexes, we only need the coordinates of the nodes of a given brain atlas. Therefore, our scripts can be adapted to different databases, image modalities, and brain atlas. A short glossary with the key terms to understand this manuscript can be found in Table 3.
Handson tutorial
General requirements
The following Python 3 packages are necessary to perform the computations presented below. The accompanying Jupyter Notebook can be found on GitHub (Table 1) or Zenodo (Centeno and Santos 2021).
Code example
The basis: the adjacency matrix
The basic unit on which graph theory and TDA are applied in the context of rsfMRI in our work is the adjacency or functional connectivity matrix (Fig. 1g), which presents the connections between all vertices in the network (Bassett and Sporns 2017; Fornito et al. 2016; Sporns 2018; Sporns et al. 2000). Typically, rsfMRI matrices are symmetric and do not specify the direction of connectivity (i.e., activity in area A drives activity in area B), thus yielding undirected networks (Fig. 1b and f). In contrast, nonsymmetric matrixes would produce directed networks.
Before calculating any metrics on such matrices, several crucial factors must be considered when dealing with connectivity data (Jalili 2016; Hallquist and Hillary 2018). One critical decision is whether one wants to keep the information about edge weights. When the edges’ weights (e.g., correlation values in rsfMRI connectivity) are maintained, the network will be weighted (Fig. 1d and f). Another approach is to use an arbitrary threshold or density, e.g., only keep and binarise the 20% strongest connections (Fig. 1a and b). There is currently no gold standard for the weighting issue in rsfMRI matrices (Fornito et al. 2016; Jalili 2016) and may also be dependent on the dataset or proposed analysis (van den Heuvel et al. 2017). Brain network data are often analysed using a specific thresholding procedure (or criteria). These thresholded brain networks often display ubiquitous signatures of the brain as a complex system, such as skewed degree distributions, clustering, Giant components, small wordness, and short average path lengths, to name a few (Eguíluz et al. 2005). However, some of these properties, considered signatures of complex networks, are observed, even when one threshold normally distributed data (Cantwell et al. (2020). Yet, some brain network properties are not robust towards changes in the threshold (Garrison et al. 2015). Those results raised the awareness for using methods of network analysis that are independent of thresholding, such as minimum spanning trees (van Dellen et al. 2018; Stam et al. 2014) or topological data analysis (Phinyomark et al. 2017), as we will discuss below. Please see Chapter 11 in Fornito et al. (2016) and van Wijk et al. (2010); Simpson et al. (2013) for a more indepth discussion on the issue of matrix thresholding and statistical connectomics.
Another relevant discussion about rsfMRI matrices is the interpretation of negative weights or anticorrelations. The debate of what such negative correlations mean in neurophysiology is still going on (Zhan et al. 2017). Studies have suggested that they could be considered artefacts introduced by global signal regression or preprocessing methods or simply by large phase differences in synchronised signals between brain areas (Chen et al. 2011; Murphy et al. 2009). Nevertheless, a few authors have suggested that anticorrelations might carry biological meaning underlying longrange synchronisation and that in diseased states, alterations in these negative correlations could indicate network reorganisation (Chen et al. 2011; Zhan et al. 2017). Negative weights can be absolutised to keep the potential biological information they may carry. If one decides to discard them, it is crucial to remember that some physiological information might be lost (Chen et al. 2011; Zhan et al. 2017; Fornito et al. 2016; Hallquist and Hillary 2018).
In this tutorial, we will use an undirected, absolutised (positively) weighted matrix. In our Jupyter notebook tutorial on GitHub (Centeno and Santos 2021), we provide an example matrix which is an average of all connectivity matrices available in our repository.
To follow the steps below, we assume that rsfMRI was already preprocessed and converted to a matrix according to some atlas. Steps and explanations on data preprocessing and atlas choices are beyond the scope of this paper; please see Strother (2006) or the NIedu course materials (Table 1) for further information. Details on our Jupyter Notebook’s dataset preprocessing can be found in Brown et al. (2012); Biswal et al. (2010).
Code example
When working with fMRI brain network data, it is helpful to generate some plots (e.g., the heatmaps for matrix visualisation and distribution plots of edge weights) to facilitate data exploration, comprehension, and flag potential artefacts. In brain networks, we expect primarily weak edges and a smaller proportion of strong ones. When plotted as a probability density of log10, we expect the weight distribution to have a Gaussianlike form Fornito et al. (2016).
Code example
Graph theory
Here, we will cover the most commonly used graph metrics in network neuroscience (see Fig. 2), also in line with Fornito et al. (2016). First, we need to create a graph object using the package NetworkX (Hagberg et al. 2008) and remove the selfloops (i.e., the connectivity matrix’s diagonal).
Code example
Degree
Vertex degree quantifies the total number of vertex connections in an undirected binary network (Fornito et al. 2016). In an undirected weighted network like our rsfMRI matrix, the vertex degree is analogous to the vertex strength (i.e., the sum of all edges of a vertex) and equivalent to its degree centrality. This metric is one of the most fundamental metrics in network analysis and is a useful summary of how densely individual vertices are connected. It can be computed as the sum of edge weights of the neighbours of vertex \(i \) as follows:
where \({w}_{ij}\) is the weight of the edge linking vertices \(i\) and \(j\).
Code example
By removing the argument weight from the function, one can compute the degree of binarised networks where all edges are either 0 or 1 (useful if working with a sparse/not fully connected matrix). This change will give the vertex degree by calculating the number of edges adjacent to the vertex. One can also remove the specified vertex to estimate the degree/strength of all vertices. The degree/strength distribution allows us to scope the general network organisation in a single shot by displaying whether the network contains a few highly connected vertices, i.e., hubs (Hallquist and Hillary 2018).
Path length

a)
The shortest path is the path with the least number of edges (or least total weight) between two vertices in a network. In a weighted graph, the shortest path is calculated by the minimum sum of the weights of edges between two vertices (Fornito et al. 2016). It is seen as a measure for understanding the efficiency of information diffusion in a network. Several algorithms can calculate path lengths, but Dijkstra’s algorithm (Dijkstra 1959) is one of the oldest and most wellknown. An important detail is that this algorithm is only applicable to graphs with nonnegative weights (Dijkstra 1959).
A pivotal point to keep in mind is that in the case of correlation matrices, such as rsfMRI data, the weights must be converted to ‘distance’ by computing the inverse of the original weight (\(1weight\) or \(\frac{1}{weight}\)); a higher correlation value represents a shorter distance (Fornito et al. 2016). This conversion is essential for all the following pathbased metrics.

b)
Average path length (or characteristic path length) is the average shortest path length for all possible pairs of vertices in a network. It is a global measure of information transport efficiency and integration in a network and is widely known due to the famous Watts–Strogatz model (Watts and Strogatz 1998). It can be computed as follows:
$$\mathrm{L }= \sum_{i,j\in \mathrm{V}} \frac{d\left(i,j\right)}{N\left(N1\right)},$$
where \(\mathrm{V}\) is a set of vertices, \(d\left(i,j\right)\) is the shortest path between vertices \(i\) and \(j\), and \(N\) is the number of vertices in the network.
Code example
The clustering coefficient
The clustering coefficient assesses the tendency for any two neighbours of a vertex to be directly connected (or more strongly connected in the weighted case) to each other and can also be termed cliquishness (Hallquist and Hillary 2018; Watts and Strogatz 1998). This metric is also used to compute the smallworldness coefficient (ratio between the characteristic path length and the clustering coefficient relative to random networks) (Watts and Strogatz 1998). The formula can be defined as follows:
where \({s}_{i}\) is the degree/strength of vertex\(i\), and the edge weights are normalised by the maximum weight in the network, such that \({\widehat{w}}_{ij} =\frac{{w}_{ij}}{\mathrm{max}\left(w\right)}\).
Code example
Centralities

a.
Eigenvector (degreebased) centrality measures a vertex’s importance in a network while also considering its neighbours’ influence (Golbeck 2013). Thus, it considers both the quantity and quality of a vertex’s connections. The eigenvector centrality can be computed from the spectra of the adjacency matrix:
$$Ax = \mathrm{\lambda x},$$where A is the adjacency matrix, and \(x\) is an eigenvector of A with eigenvalue λ. We can now define the eigenvector centrality of a vertex \(i\) as the following sum over its neighbours:
$${C}_{E}\left(\mathrm{i}\right) = \frac{1}{{\uplambda }_{1}} \sum_{j=1}^{N} {A}_{ij}{x}_{j.}$$For weighted networks, certain conditions apply. According to the Perron–Frobenius theorem, the adjacency matrix’s largest eigenvalue, denoted here by \({\lambda }_{1}\), must be unique and positive, guaranteed only for matrices with positive values (Fornito et al. 2016; Newman 2008).

b.
Closeness (shortest pathbased) centrality measures how closely or’ directly’ connected a vertex is to the rest of the network. If the vertex is the closest to every other element in the network, it has the potential to spread information fast and efficiently (Fornito et al. 2016). Formally, the closeness centrality of a vertex \(i\) is the inverse of its average shortest path length (N.B. weights need to be converted to distances) to all N − 1 other vertices:
$$ Cc(i) = \frac{N  1}{{\sum\nolimits_{i = 1}^{N  1} {d(i,\,j)} }}, $$where d(\(i\), \(j\)) is the shortestpath distance between \(i\) and \(j\), and \(N\) is the number of vertices in the graph. In weighted networks, closeness centrality can be estimated by considering the summed weight of the shortest paths according to Dijkstra’s algorithm (Dijkstra 1959).

c.
Betweenness (shortest pathbased) centrality is the proportion of all vertexpairs shortest paths in a network that pass through a particular vertex (Newman 2008; Freeman 1977). It is used to understand the influence of vertices in the overall flow of information in a network. To compute the betweenness centrality of a vertex \(i\), one has to calculate the proportion of shortest paths between two vertices, e.g., \(i, j\), that pass through vertex \(h\):
$${\mathrm{C}}_{B}\left(h\right)= \sum_{i\ne h\ne j\in \mathrm{V}}\frac{{\sigma }_{ij}\left(h\right)}{{\sigma }_{ij}} ,$$
where \(V\) is a set of vertices, \({\sigma }_{ij}\) is the total number of shortest paths between \(i\) and \(j\), and \({\sigma }_{ij}\left(h\right)\) is the number of those paths that pass through \(h\). For weighted graphs, edges must be greater than zero, and the metric considers the sum of the weights (Fornito et al. 2016). Again, it is necessary to use the distance when using this shortest pathbased metric. This formula can also be normalised by putting \(\frac{2}{\left(N1\right)\left(N1\right)}\) in front of the sum (N being the number of vertices).
Code example
The minimum spanning tree
The minimum spanning tree is the backbone of a network, i.e., the minimum set of edges necessary to ensure that paths exist between all vertices without forming cycles (Stam et al. 2014; van Dellen et al. 2018). A few main algorithms are used to build the spanning tree, with Kruskal’s algorithm being implemented in NetworkX (Kruskal 1956). Briefly, this algorithm ranks the distances between vertices, adds the ones with the smallest distance first, and at each added edge, it checks if cycles are formed or not. The algorithm will not keep an edge that results in the formation of a cycle.
Code example
Modularity
Modularity states how divisible a network is into different modules (or communities). The identification of the modules is performed by the community detection algorithm (Fornito et al. 2016; Meunier et al. 2010; Bullmore and Sporns 2009). Here, we will use the Louvain algorithm (Blondel et al. 2008) as recommended by Fornito et al. (2016). It works in a twostep iterative manner, first looking for communities by optimising modularity locally and then concatenating vertices that belong to the same module (Blondel et al. 2008).
Code example
Topological data analysis
In this section, we will use TDA on our rsfMRI adjacency matrices. TDA can identify different network characteristics by addressing the highorder structure of a network beyond pairwise connections as used in graph theory (Carlsson 2020; Battiston et al. 2020; KartunGiles and Bianconi 2019). TDA generally uses topology and geometry methods to study the shape of the data (Carlsson 2009). A core feature of TDA is the ability to provide robust results compared with alternative methods, even if the data are noisy (Blevins and Bassett 2020; Expert et al. 2019). In this context, it is essential to frame the difference between noise and systematic error for applications of TDA properly. Noise is caused by factors that affect the measurement of the variable of interest entirely at random. Systematic errors, however, are not determined exclusively by chance. They are introduced by a factor that systematically influences the variable of interest measurement (e.g. by an inaccuracy involving either the observation or measurement process). In rsfMRI (and other brainrelated measures), both types of noise can be present.
One of the benefits of using TDA in network neuroscience is the possibility of finding global properties of a network that are preserved regardless of the way we represent the network (Petri et al. 2014), as we will illustrate below. Those properties are the socalled topological invariants.
We will cover a few fundamental TDA concepts: filtration, simplicial complexes, Euler characteristic, phase transitions, Betti numbers, curvature, and persistent homology. A summary can be found in Fig. 3.
The basis: the adjacency matrix and filtration
As indicated in the earlier section on graph theory, there is no consensus on the necessity or level of thresholding performed on rsfMRIbased adjacency matrices. However, TDA overcomes this problem by investigating functional connectivity over all possible thresholds in a network. This process of investigating network properties looking for all possible thresholds instead of choosing a fixed one is called filtration (Fig. 3b and Supplementary Material 1). It consists of changing the threshold, e.g., the density d of the network, from \(0 \le d\le 1\). This yields a nested sequence of networks, in which increasing d leads to a more densely connected network. Notice that the notion of filtration is not only used in high order interactions but has also been applied in pair wise, graphtheoretical work (Wang et al. 2018; GraciaTabuenca et al. 2020).
Code example
Simplicial complexes
In TDA, we consider that the network as a multidimensional structure called the simplicial complex. Such a network is not only made up of the set of vertices (0simplex) and edges (1simplex) but also of triangles (2simplex), tetrahedrons (3simplex), and higher kdimensional structures (Fig. 3a). In short, a ksimplex is an object in kdimensions and, in our work, is formed by a subset of k+1 vertices of the network (Fig. 4).
We can encode a network into a simplicial complex in several ways (Lambiotte et al. 2019; Edelsbrunner and Harer 2010; Maletić et al. 2008). However, here, we will focus on building a simplicial complex only from the brain network’s cliques, i.e., we will create the socalled clique complex of a brain network. In a network, a kclique is a subset of the network with \(k\) alltoall connected nodes. 0clique corresponds to the empty set, 1cliques correspond to nodes, 2cliques to links, 3cliques to triangles, and so on. In the clique complex, each k + 1 clique is associated with a ksimplex. This choice for creating simplexes from cliques has the advantage of using pairwise signal processing to create a simplicial complex from brain networks, such as in Giusti et al. (2015). It is essential to mention that other strategies to build simplicial complexes beyond pairwise signal processing are still under development, such as applications using multivariate information theory together with tools from algebraic topology (Baudot 2019a, b; Barbarossa and Sardellitti 2020; Baudot et al. 2019; Baudot and Bennequin 2015; Rosas et al. 2019; Gatica et al. 2020).
Our Jupyter Notebook provides the code to visualise the clique complex developed in (Santos et al. 2019). To create the 3D plots, we used mesh algorithms available in Plotly (Inc. 2015), together with a mesh surface of the entire brain available in Fan et al. (2016); Bakker et al. (2015). In Fig. 5, we display an example of 3D visualisation of 3cliques in the 1000 Functional Connectomes data. When we increase the filtration density d, we obtain more connections, and more 3cliques arise. In Fig. 5, only 3cliques are shown; however, the same can be done for higherdimensional cliques like a tetrahedron, et cetera. In Supplementary Material 1, we offer filtration in a functional brain network up to 4cliques. In the Jupyter Notebook, we can also visualise the clique complex at arbitrary sizes, up to computational limits. The computation is not shown here as code blocks due to its size and complexity (see the Jupyter Notebook).
The Euler characteristic
The Euler characteristic is one example of topological invariants: the network properties that do not depend on a specific graph representation. We first introduce the Euler characteristic for polyhedra, as illustrated in Fig. 6. Later, we translate this concept to brain networks. In 3D convex polyhedra (for example, a cube, a tetrahedron, et cetera, see Fig 6), the Euler characteristic is defined as the numbers of vertices minus edges plus faces of the considered polyhedra. For convex polyhedra without cavities (holes in its shape), which are isomorphous to the sphere, the Euler characteristic is always two, as you can see in Fig. 6. If we take the cube and make a cavity, the Euler drops to zero as it is in the torus. If we make two cavities in a polyhedral (as in the bitorus), the Euler drops to minus two (Fig. 7). We can understand that the Euler characteristic tells us something about a polyhedron’s topology and its analogous surface. In other words, if we have a surface and we make a discrete representation of it (e.g., a surface triangulation), its Euler characteristic will always be the same, regardless of the way we do it.
We can now generalise the definition of Euler characteristic to simplicial complex in any dimension. Thus, the high dimensional version of the Euler characteristic is expressed by the alternate sum of the numbers \({Cl}_{k }\left(d\right)\) of the kcliques (which are (k1)simplexes) present in the network’s simplicial complex for a given value of the density threshold d.
Code example
Note that the clique algorithm, the primary function used in our code (euler—Supplementary Material 2), is an NPcomplete problem, which is computationally expensive for large and/or dense networks, regardless of how you implement it (Pardalos and Xue 1994). An alternative is to fix an upper bound for the cliques’ size (Pardalos and Xue 1994; Gillis 2018). Therefore, the second function (euler_k—Supplementary Material 2) allows the user to constrain the maximum size of the cliques we are looking for. This means that we are fixing the dimension k of our simplicial complex and ignoring simplexes of dimension greater than k.
Topological phase transitions
Phase transitions can provide insight into the proprieties of a ‘material’. For example, water is known for becoming steam at 100 °C. Similarly, by using TDA when comparing a patient and healthy population, one could identify these populations’ properties by studying each group’s topological phase transition profile. This strategy has already been applied for investigating group differences between controls and glioma brain networks (Santos et al. 2019) and typically developing children and children with attentiondeficit/hyperactivity disorder (GraciaTabuenca et al. 2020). In other fields, topological phase transitions were also investigated in the S. cerevisiae and C. elegans protein interaction networks, reionisation processes, and evolving coauthorship networks (Amorim et al. 2019; Giri and Mellema 2021; Lee et al. 2021).
To investigate topological phase transitions in brain networks, we first need to visualise the Euler entropy (Fig 3 in Santos et al. 2017):
\({S}_{\upchi }\)= ln\(\upchi \) .when \(\upchi \) = 0 for a given value of the density of the network, the Euler entropy is singular, \({S}_{\upchi } \to \infty \). Under specific hypotheses, a topological phase transition in a complex network occurs when the Euler characteristic is null (Santos et al. 2019). This statement finds support in the behaviour of \({S}_{\upchi }\) at the thermodynamic phase transitions across various physical systems (Santos et al. 2017). In network theory, the Giant component transition is associated with network changes, from smaller connected clusters to the emergence of Giant ones (Erdős 1959). Theoretically, topological phase transitions are related to the extension of the Giant component transition for simplicial complexes (Linial and Peled 2016). Based on numerical simulations, it was also conjectured that the longest cycle is born in the phase transition vicinity (Bobrowski and Skraba 2020; Speidel et al. 2018). Phase transitions can also be visualised in Birth/Death plots (Fig. 3e) which will be discussed later in the Persistent Homology section.
Betti numbers
Another set of topological invariants are the Betti numbers (\(\beta \)). Given that a simplicial complex is a highdimensional structure, \({\beta }_{k}\) counts the number of kdimensional holes in the simplicial complex. These are topological invariants that correspond, for each \(k\ge \) 0, to the number of linearly independent kdimensional holes in the simplicial complex (Zomorodian 2005).
In Fig. 8, we show the representation of the kdimensional holes. We give one example for each dimension. In a simplicial complex, there can be many of these kholes and counting them provide the Betti number \(\beta \), e.g., if \({\beta }_{2}\) is equal to five, there are 5 twodimensional holes.
Code example
Notice that for higher k or dense simplicial complexes, the calculation of the \(\beta \) becomes computationally expensive.
There are ways to estimate the \(\beta \) of a simplicial complex without calculating it directly. It is known that the \(\beta \) relate to Euler characteristics and phase transitions. The Euler characteristics of a simplicial complex can also be computed using the \(\beta \) via the following formula (Edelsbrunner and Harer 2010):
where \(k\_\mathrm{max}\) is the maximum dimension that we are computing the cycles.
Furthermore, topological phase transitions are also defined as the \(\beta \) of a simplicial complex (Bobrowski and Kahle 2018). We know that \({\upbeta }_{0}\) counts the number of the connected components of a simplicial complex. Suppose we compute the Betti curves as a function of probability in stochastic models. In that case, each Betti curve passes through two distinct phases in a narrow interval: one when it first emerges and the other when it vanishes (Linial and Peled 2016). That means that, under similar assumptions as in theoretical models, if the \(\beta \) distribution is unimodal, increasing the density of edges of a brain network will lead to the appearance of \(\beta \) of a higher order. In contrast, smaller Betti numbers will disappear at the vicinity of a topological phase transition.
In Fig. 9, we illustrate this property on simplicial complexes obtained from random networks. As the probability increases and so the density of the network is higher, we find a sequence of dominant \({\beta }_{k}\) starting from k = 0, that change (i.e., k is incremented by one unity) every time a topological phase transition occurs. While the singularities of the Euler entropy \({S}_{\upchi }\) determine the transitions’ location, the crossover of the \(\beta \) characterises which kind of multidimensional hole prevails in each topological phase of the filtration.
Curvature
Curvature is a TDA metric that can link the global network properties described above to local features (Weber et al. 2017; Farooq et al. 2019; Santos et al. 2019). When working with brain network data, this will allow us to compute topological invariants for the wholebrain set of vertices and understand the contribution of specific individual nodal, or subnetwork, geometric proprieties to global properties of the brain network.
Several approaches to defining a curvature for networks are available (Najman et al. 2017; Weber et al. 2017), including some already used in neuroscientific investigations (Santos et al. 2019). We will illustrate the curvature approach linked to topological phase transitions, previously introduced for complex systems in (Farooq et al. 2019; Najman et al. 2017; Wu et al. 2015).
To compute the curvature (Supplementary Material 4), filtration is used to calculate the clique participation rank (i.e., the number of \(k\)cliques in which a vertex \(i\) participates for density d) (Sizemore et al. 2018), which we denote here by \({Cl}_{ik }\left(d\right)\). The curvature of the vertex based on the participation rank is then defined as follows:
where \({Cl}_{ik}\) = 1 since each vertex \(i\) participates in a single 1clique (the vertex itself), and \({k}_{max}\) the maximum number of vertices that are alltoall connected in the network.
To link this nodal curvature to the network’s global properties, we use the GaussBonnet theorem for networks, through which one can connect a local curvature of a network and its Euler characteristic. Conversely, by summing up all the curvatures of the network across different thresholds, one can reach the alternate sum of the numbers \(C{l}_{k}\) of kcliques (a subgraph with k alltoall connected vertices) present in the simplicial complex of the network for a given density threshold d ∈ [0, 1], according to the following equation:
By doing so, we also write the Euler characteristics as a sum of the curvature of all network vertices, i.e.,
We illustrate the curvature distribution for a functional brain network for densities before and after the transition in Fig. 10.
Persistent homology
Homology is a topology branch that investigates objects’ shapes by studying their holes (or cycles). Persistent homology tracks the emergence of cycles across the evolving simplicial complexes during filtration, allowing us to recognise whether there were homology classes that “persisted” for many filtrations (time here meaning the threshold gap between the birth and death of a cycle) (Curto 2017; Giusti et al. 2016). Importantly, to compute persistent homology, we need to work with a distance matrix, the first step in the code below. We can then calculate the simplicial complex’s persistence and plot it as a barcode or a persistence diagram (Fig. 3c and e). Here we used the Gudhi package for the implementation of those steps (Maria et al. 2014). The topological phase transitions in complex networks (Amorim et al. 2019; Santos et al. 2019) can also be identified between the changes in the dimensionality of the birth/death graphs mentioned above (Fig. 3e).
Code example
Discussion
This tutorial has explained some of the main metrics related to two network neuroscience branches—graph theory and TDA—providing short theoretical backgrounds and code examples accompanied by a publicly available Jupyter Notebook. We innovate by combining handson explanations with readytouse codes of these subfields and visualisations of simplicial complexes in the brain, hopefully lowering the high threshold necessary for neuroscientists to get acquainted with these new analysis methods, particularly for these new methods rsfMRI data. Here, we also innovate by providing realistic visualisation of higherorder simplices in brain networks.
Our main goal was to provide a stepbystep computational tutorial to use graph theory and TDA on brain imaging data, particularly rsfMRI, with indepth explanations behind each metric. The core idea of applying these analysis frameworks to brain data is that both frameworks can quantitatively combine two evidently essential characteristics of the brain: the brain not only works both at a local level in specialised brain regions but also contains apparent global properties that are of importance for its functioning, which are usually investigated in isolation. As a potentially powerful fusion between localizationism and holism, graph theory and TDA concepts have already been applied in brain research. Starting with graph theory, all the metrics mentioned above have been used in the investigation of brain networks in both normal or pathological states (Eijlers et al. 2017; GarciaGarcia et al. 2015; Wang et al. 2017; Wink 2019; Breedt et al. 2021; DeSalvo et al. 2020; Liu et al. 2012; dos Santos Siqueira et al. 2014; Yu et al. 2012; Davis et al. 2013; Suo et al. 2015). As one can identify by reading these articles, researchers often use different graphtheoretical metrics in the same study, which helps them look for alterations that might explain group differences in specific contexts (gender, age, pathology, development). This brief review and commentary (Eijlers et al. 2019) summarise some applications. Now, moving on to the newer framework of TDA in neuroscience, fewer studies have been published using rsfMRI data. Santos et al. (2019) applied the concepts of the Euler characteristic, topological phase transitions and curvature in human brain data, to show that these transitions can be found in brain data, helping pave the way for TDA in brain data applications.
Moreover, alterations in wholebrain connectomes were identified in attentiondeficit/hyperactivity disorder subjects using Betti numbers and persistent homology, complementing connectomicsrelated methods that aim to identify the markers of this disorder (GraciaTabuenca et al. 2020). A similar approach was used in an Alzheimer’s disease dataset by Kuang et al. (2019). More considerations on how TDA can be used in brain imaging big data and restingstate functional connectivity analyses can be found in Phinyomark et al. (2017); Petri et al. (2014); Anderson et al. (2018); Saggar et al. (2018); Salch et al. (2021); Songdechakraiwut and Chung (2020).
Notably, limitations and other relevant points should be kept in mind when working with these metrics. Firstly, it is common in network neuroscience to use null models for comparison with real data. The idea is to show that the results are different from what one would obtain by chance (or randomly). The generation and comparison with null models must be performed differently for graph theory and TDA, and it is crucial to define what propriety should be kept constant (e.g., the density of the network or degree distribution). For instance, in Viger and Latapy (2005), if one wants to generate null models with a prescribed degree sequence. In this context, simplicial complexes built from ErdoRenyi networks illustrated in Fig. 9 are the simplest (and by no means realistic) null models we can generate.
Nevertheless, the computation and discussion of null models are beyond this tutorial’s scope and would be an article in itself. A more indepth discussion of null models in graph theory can be found in Fornito et al. (2016). Please see Sect. 4 of Battiston et al. (2020) and Blevins and Bassett (2020) for null models in simplicial complexes.
Moreover, it is crucial to appreciate limitations in interpretation when using these metrics in connectivitybased data. Since rsfMRI data is often calculated as a temporal correlation between time series using Pearson’s correlation coefficient, a bias on the number of triangles can emerge. For example, suppose areas A and B and areas C and B are communicating and thus correlated. In that case, a correlation will be present between A and C, even if there would be no actual communication between these vertices (Zalesky et al. 2012). This can affect graphtheoretical metrics such as the clustering coefficient, with networks based on this statistical method being automatically more clustered than random models, and TDA metrics, where the impact depends on how highorder interactions are defined. The proper way to determine and infer highorder interactions in the brain is an ongoing challenge in network neuroscience. Here we simplified our approach using the cliques of a network to define our simplicial complex. For those interested in a more indepth discussion on the topic, we recommend Sects. 1 and 3 of chapters 7 and 10, respectively, in Fornito et al. (2016).
The use of weighted matrices can also come with caveats. As mentioned above, various metrics use the sum of weights to compute final nodal values. From that, multiple edges with low weights might have a final sum equal to a few edges with higher weights. How to deal with this limitation and distinguish between these cases is still under discussion. A possible solution was proposed by Opsahl et al. (2010), in which the addition of a tunable parameter in the computation of centralities can allow the researcher to include the number of edges in the total sum, not only the sum of the weights.
Concerning TDA, it is essential to think about limitations in its use due to computational power. The computation of cliques falls in the cliqueproblem, an NP (nonpolynomial time) problem, thus listing cliques may require exponential time as the size of the cliques or networks grows (Gillis 2018; Pardalos and Xue 1994). For example, if the matrix to be analysed has 60 vertices with a maximum clique size of 23, this will correspond to \(\sum \left(\genfrac{}{}{0pt}{}{60}{k}\right)\) for \(k \in \left\{0,\dots , 23\right\}\) cliques, resulting in an enormous amount of time to compute all cliques. What we can do for practical applications is to limit the clique size that can be reached by the algorithm, which determines the dimension of the simplicial complex in which the brain network is represented. This arbitrary constraint implies a theoretical simplification, limiting the space or the dimensionality in which we would analyse brain data. Another issue is that, to finish TDA computations in a realistic timeframe, the researcher might need to establish a maximal threshold/density for convergence even after reducing the maximal clique size. Even though TDA approaches lead to substantial improvements in network science; apart from applications using the Mapper algorithm (Saggar et al. 2018), the limitations mentioned above contribute to losing information on the data’s shape (Stolz 2014).
Furthermore, given the early stage of TDA approaches in clinical network neuroscience, it is relevant to recognise that the neurobiological meaning of the metrics mentioned here is still limited. Further studies contrasting different neuroscientific techniques with TDA must be done to improve the understanding, in the neurobiological level, on what a topological metrics represent and how they correlate with brain functioning. However, it is already possible to use these metrics to differentiate groups (Santos et al. 2019; GraciaTabuenca et al. 2020), and plausible to assume that the interpretation of some classical metrics could be extrapolated to higher orders interactions. For example, the concept of the centralities using pairwise interactions is used to understand node importance and hubs, the same, in theory, could be applied to the relationships between 3 or more vertices by extending the definition of centrality from networks to simplicial complexes, as done in Hernández Serrano and Sánchez Gómez (2020); Estrada and Ross (2018).
Last, we would like to briefly mention more general problems in network neuroscience and brain imaging. Before applying graph theoretical or topological data analysis, one should be aware of frequent arbitrary decisions such as defining thresholds, using binary or weighted matrices, and controlling for density. Besides, one should think about the differences that arise from using particular atlases and parcellations and their influence on the findings (Wang et al. 2009; Douw et al. 2019; Fornito et al. 2016; GraciaTabuenca et al. 2020; Wu et al. 2019; Eickhoff et al. 2018; Bullmore and Sporns 2009). All these factors can impact how credible and reproducible the field of network neuroscience will be, inevitably influencing how appealing the metrics’ use might be to clinical practice (Douw et al. 2019).
Conclusion
Network neuroscience is pivotal in the understanding of brain organisation and function. Graph theory has been the most utilised framework so far, but as the field of network neuroscience expands, newer methods such as TDA are starting to take part in the investigation. To further improve the field, especially in clinical network neuroscience, it is imperative to make the computation of the developed metrics accessible, easy to comprehend, visualise, and efficient. Moreover, researchers must be aware of the crucial decisions one must make when executing data analysis and how these can affect studies’ results and reproducibility. We hope to have facilitated the comprehension of some aspects of network and topological neuroscience, the computation and visualisation of some of its metrics. As a final reminder, we would again suggest the reader to explore our table of resources and the Jupyter Notebook developed by our team.
Availability of data and material
Not applicable.
Notes
Notice that the notion of metric in mathematics defines distance between two points in a set (Do Carmo 2016), which is distinct from what we are using in this work. We denote as metric any quantity that can be computed, i.e., “measured”, in a brain network or simplicial complex.
References
Amorim E, Moreira RA, Santos FAN (2019) The Euler characteristic and topological phase transitions in complex systems. BioRxiv. https://doi.org/10.1101/871632
Anderson KL, Anderson JS, Palande S, Wang B (2018) Topological data analysis of functional MRI connectivity in time and space domains. In: Wu G, Rekik I, Schirmer MD, Chung AW, Munsell B (eds) International workshop on connectomics in neuroimaging, Granada, Spain. Connectomics in neuroimaging. Springer, Cham, pp 67–77
Bakker R, Tiesinga P, Kötter R (2015) The scalable brain atlas: instant webbased access to public brain atlases and related content. Neuroinformatics 13(3):353–366. https://doi.org/10.1007/s120210149258x
Barbarossa S, Sardellitti S (2020) Topological signal processing over simplicial complexes. IEEE Trans Signal Process 68:2992–3007. https://doi.org/10.1109/TSP.2020.2981920
Bassett DS, Bullmore ET (2017) Smallworld brain networks revisited. Neuroscientist 23(5):499–516. https://doi.org/10.1177/1073858416667720
Bassett DS, Sporns O (2017) Network neuroscience. Nat Neurosci 20(3):353. https://doi.org/10.1038/nn.4502
Battiston F, Cencetti G, Iacopini I, Latora V, Lucas M, Patania A, Young JG, Petri G (2020) Networks beyond pairwise interactions: structure and dynamics. Phys Rep 874:1–92. https://doi.org/10.1016/j.physrep.2020.05.004
Baudot P (2019a) Elements of qualitative cognition: an information topology perspective. Phys Life Rev 31:263–275. https://doi.org/10.1016/j.plrev.2019.10.003
Baudot P (2019b) The PoincareShannon machine: statistical physics and machine learning aspects of information cohomology. Entropy 21(9):881. https://doi.org/10.3390/e21090881
Baudot P, Bennequin D (2015) The homological nature of entropy. Entropy 17(5):3253–3318. https://doi.org/10.3390/e17053253
Baudot P, Tapia M, Bennequin D, Goaillard JM (2019) Topological information data analysis. Entropy 21(9):869. https://doi.org/10.3390/e21090869
Biswal BB, Mennes M, Zuo XN, Gohel S, Kelly C, Smith SM, Beckmann CF, Adelstein JS, Buckner RL, Colcombe S, Dogonowski AM, Ernst M, Fair D, Hampson M, Hoptman MJ, Hyde JS, Kiviniemi VJ, Kotter R, Li SJ, Lin CP, Lowe MJ, Mackay C, Madden DJ, Madsen KH, Margulies DS, Mayberg HS, McMahon K, Monk CS, Mostofsky SH, Nagel BJ, Pekar JJ, Peltier SJ, Petersen SE, Riedl V, Rombouts SA, Rypma B, Schlaggar BL, Schmidt S, Seidler RD, Siegle GJ, Sorg C, Teng GJ, Veijola J, Villringer A, Walter M, Wang L, Weng XC, WhitfieldGabrieli S, Williamson P, Windischberger C, Zang YF, Zhang HY, Castellanos FX, Milham MP (2010) Toward discovery science of human brain function. Proc Natl Acad Sci USA 107(10):4734–4739. https://doi.org/10.1073/pnas.0911855107
Blevins AS, Bassett DS (2020) Reorderability of nodefiltered order complexes. Phys Rev E 101(5–1):052311. https://doi.org/10.1103/PhysRevE.101.052311
Blevins AS, Kim JZ, Bassett DS (2021) Variability in higher order structure of noise added to weighted networks. https://www.nature.com/articles/s4200502100725x
Blondel VD, Guillaume JL, Lambiotte R (2008) Fast unfolding of communities in large networks. J Stat Mech: Theory Exp 10:P10008. https://doi.org/10.1088/17425468/2008/10/p10008
Bobrowski O, Kahle M (2018) Topology of random geometric complexes: a survey. J Appl Comput Topol 1(3):331–364. https://doi.org/10.1007/s4146801700100
Bobrowski O, Skraba P (2020) Homological percolation and the Euler characteristic. Phys Rev E 101(3):032304. https://doi.org/10.1103/PhysRevE.101.032304
Breedt LC, Santos FAN, Hillebrand A, Reneman L, van Rootselaar AF, Schoonheim MM, Stam CJ, Ticheler A, Tijms BM, Veltman DJ, Vriend C, Wagenmakers MJ, van Wingen GA, Geurts JJG, Schrantee A, Douw L (2021) Multimodal multilayer network centrality relates to executive functioning. BioRxiv. https://doi.org/10.1101/2021.06.28.450180
Brown JA, Rudie JD, Bandrowski A, Van Horn JD, Bookheimer SY (2012) The UCLA multimodal connectivity database: a webbased platform for brain connectivity matrix sharing and analysis. Front Neuroinform 6:28. https://doi.org/10.3389/fninf.2012.00028
Bullmore E, Sporns O (2009) Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 10(3):186–198. https://doi.org/10.1038/nrn2575
Cantwell GT, Liu Y, Maier BF, Schwarze AC, Serván CA, Snyder J, StOnge G (2020) Thresholding normally distributed data creates complex networks. Phys Rev E 101(6):062302. https://doi.org/10.1103/PhysRevE.101.062302
Carlsson G (2009) Topology and data. Bull Am Math Soc 46(2):255–308. https://doi.org/10.1090/S027309790901249X
Carlsson G (2020) Topological methods for data modelling. Nat Rev Phys 2(12):697–708. https://doi.org/10.1038/s42254020002493
Centeno EGZ, Santos FN (2021) Notebook for network and topological analysis in neuroscience. Zenodo. https://doi.org/10.5281/zenodo.4483651
Chen G, Chen G, Xie C, Li SJ (2011) Negative functional connectivity and its dependence on the shortest path length of positive network in the restingstate human brain. Brain Connect 1(3):195–206. https://doi.org/10.1089/brain.2011.0025
Cowan WM, Harter DH, Kandel ER (2000) The emergence of modern neuroscience: some implications for neurology and psychiatry. Annu Rev Neurosci 23:343–391. https://doi.org/10.1146/annurev.neuro.23.1.343
Curto C (2017) What can topology tell us about the neural code? Bull Am Math Soc 54(1):63–78. https://doi.org/10.1090/bull/1554
Curto C, Itskov V (2008) Cell groups reveal structure of stimulus space. PLoS Comput Biol 4(10):e1000205. https://doi.org/10.1371/journal.pcbi.1000205
Davis FC, Knodt AR, Sporns O, Lahey BB, Zald DH, Brigidi BD, Hariri AR (2013) Impulsivity and the modular organization of restingstate neural networks. Cereb Cortex 23(6):1444–1452. https://doi.org/10.1093/cercor/bhs126
DeSalvo MN, Tanaka N, Douw L, Cole AJ, Stufflebeam SM (2020) Contralateral Preoperative RestingState Functional MRI network integration is associated with surgical outcome in temporal lobe epilepsy. Radiology 294(3):622–627. https://doi.org/10.1148/radiol.2020191008
Dijkstra EW (1959) A note on two problems in connexion with graphs. Numer Math 1(1):269–271. https://doi.org/10.1007/BF01386390
Do Carmo MP (2016) Differential geometry of curves and surfaces: revised and updated, 2nd edn. Courier Dover Publications, Mineola
dos Santos SA, Biazoli Junior CE, Comfort WE, Rohde LA, Sato JR (2014) Abnormal functional restingstate networks in ADHD: graph theory and pattern recognition analysis of fMRI data. Biomed Res Int 2014:380531. https://doi.org/10.1155/2014/380531
Douw L, van Dellen E, Gouw AA, Griffa A, de Haan W, van den Heuvel M, Hillebrand A, Van Mieghem P, Nissen IA, Otte WM, Reijmer YD, Schoonheim MM, Senden M, van Straaten ECW, Tijms BM, Tewarie P, Stam CJ (2019) The road ahead in clinical network neuroscience. Netw Neurosci 3(4):969–993. https://doi.org/10.1162/netn_a_00103
Edelsbrunner H, Harer J (2010) Computational topology: an introduction, vol 69, 1st edn. American Mathematical Society, Providence
Eguíluz VM, Chialvo DR, Cecchi GA, Baliki M, Apkarian AV (2005) Scalefree brain functional networks. Phys Rev Lett 94(1):018102. https://doi.org/10.1103/PhysRevLett.94.018102
Eickhoff SB, Yeo BTT, Genon S (2018) Imagingbased parcellations of the human brain. Nat Rev Neurosci 19(11):672–686. https://doi.org/10.1038/s4158301800717
Eijlers AJ, Meijer KA, Wassenaar TM, Steenwijk MD, Uitdehaag BM, Barkhof F, Wink AM, Geurts JJ, Schoonheim MM (2017) Increased defaultmode network centrality in cognitively impaired multiple sclerosis patients. Neurology 88(10):952–960. https://doi.org/10.1212/WNL.0000000000003689
Eijlers AJC, Wink AM, Meijer KA, Douw L, Geurts JJG, Schoonheim MM (2019) Functional network dynamics on functional MRI: a primer on an emerging frontier in neuroscience. Radiology 292(2):460–463. https://doi.org/10.1148/radiol.2019194009
Erdős PR (1959) On random graph. Publ Math 6:290–297
Estrada E, Ross GJ (2018) Centralities in simplicial complexes. Applications to protein interaction networks. J Theor Biol 438:46–60. https://doi.org/10.1016/j.jtbi.2017.11.003
Expert P, Lord LD, Kringelbach ML, Petri G (2019) Editorial: topological neuroscience. Netw Neurosci 3(3):653–655. https://doi.org/10.1162/netn_e_00096
Fan L, Li H, Zhuo J, Zhang Y, Wang J, Chen L, Yang Z, Chu C, Xie S, Laird AR, Fox PT, Eickhoff SB, Yu C, Jiang T (2016) The human brainnetome atlas: a new brain atlas based on connectional architecture. Cereb Cortex 26(8):3508–3526. https://doi.org/10.1093/cercor/bhw157
Farahani FV, Karwowski W, Lighthall NR (2019) Application of graph theory for identifying connectivity patterns in human brain networks: a systematic review. Front Neurosci 13:585. https://doi.org/10.3389/fnins.2019.00585
Farooq H, Chen Y, Georgiou TT, Tannenbaum A, Lenglet C (2019) Network curvature as a hallmark of brain structural connectivity. Nat Commun 10(1):4937. https://doi.org/10.1038/s4146701912915x
Fornito A, Zalesky A, Bullmore E (2016) Fundamentals of brain network analysis, 1st edn. Academic Press, San Diego
Freeman LC (1977) A set of measures of centrality based on betweenness. Sociometry 40(1):35–41. https://doi.org/10.2307/3033543
GarciaGarcia I, Jurado MA, Garolera M, MarquesIturria I, Horstmann A, Segura B, Pueyo R, SenderPalacios MJ, VernetVernet M, Villringer A, Junque C, Margulies DS, Neumann J (2015) Functional network centrality in obesity: a restingstate and task fMRI study. Psychiatry Res 233(3):331–338. https://doi.org/10.1016/j.pscychresns.2015.05.017
Garrison KA, Scheinost D, Finn ES, Shen X, Constable RT (2015) The (in)stability of functional brain network measures across thresholds. Neuroimage 118:651–661. https://doi.org/10.1016/j.neuroimage.2015.05.046
Gatica M, Cofré R, Mediano PAM, Rosas FE, Orio P, Diez I, Swinnen SP, Cortes JM (2020) Highorder interdependencies in the aging brain. BioRxiv. https://doi.org/10.1101/2020.03.17.995886
Gillis A (2018) The clique problem—a polynomial time and nonheuristic solution. viXra. https://doi.org/10.13140/RG.2.2.14191.07841
Giri SK, Mellema G (2021) Measuring the topology of reionization with Betti numbers. https://www.divaportal.org/smash/record.jsf?pid=diva2%3A1596056&dswid=180
Giusti C, Pastalkova E, Curto C, Itskov V (2015) Clique topology reveals intrinsic geometric structure in neural correlations. Proc Natl Acad Sci USA 112(44):13455–13460. https://doi.org/10.1073/pnas.1506407112
Giusti C, Ghrist R, Bassett DS (2016) Two’s company, three (or more) is a simplex: algebraictopological tools for understanding higherorder structure in neural data. J Comput Neurosci 41(1):1–14. https://doi.org/10.1007/s1082701606086
Golbeck J (2013) Chapter 3—network structure and measures. In: Golbeck J (ed) Analyzing the social web. Morgan Kaufmann, Boston, pp 25–44
GraciaTabuenca Z, DiazPatino JC, Arelio I, Alcauter S (2020) Topological data analysis reveals robust alterations in the wholebrain and frontal lobe functional connectomes in attentiondeficit/hyperactivity disorder. eNeuro. https://doi.org/10.1523/ENEURO.054319.2020
Greve DN, Brown GG, Mueller BA, Glover G, Liu TT, Function Biomedical Research N (2013) A survey of the sources of noise in fMRI. Psychometrika 78:396–416. https://doi.org/10.1007/s1133601292940
Gross JL, Yellen J (2003) Handbook of graph theory, 1st edn. CRC Press, Bosa Roca
Hagberg A, Swart P, Chult SD (2008) Exploring network structure, dynamics, and function using NetworkX. In: G Varoquaux TV, J Millman (ed) Proceedings of the 7th Python in Science conference (SciPy 2008), Pasadena, USA, Aug 19–24. p 11–15
Hallquist MN, Hillary FG (2018) Graph theory approaches to functional network organization in brain disorders: a critique for a brave new smallworld. Netw Neurosci 3(1):1–26. https://doi.org/10.1162/netn_a_00054
Hart MG, Ypma RJ, RomeroGarcia R, Price SJ, Suckling J (2016) Graph theory analysis of complex brain networks: new concepts in brain mapping applied to neurosurgery. J Neurosurg 124(6):1665–1678. https://doi.org/10.3171/2015.4.JNS142683
Hernández Serrano D, Sánchez Gómez D (2020) Centrality measures in simplicial complexes: applications of topological data analysis to network science. Appl Math Comput 382:125331. https://doi.org/10.1016/j.amc.2020.125331
Jalili M (2016) Functional brain networks: does the choice of dependency estimator and binarization method matter? Sci Rep 6:29780. https://doi.org/10.1038/srep29780
KartunGiles AP, Bianconi G (2019) Beyond the clustering coefficient: a topological analysis of node neighbourhoods in complex networks. Chaos Solitons Fractals: X 1:100004. https://doi.org/10.1016/j.csfx.2019.100004
Kruskal JB (1956) On the shortest spanning subtree of a graph and the traveling salesman problem. Proc Am Math Soc 7(1):48–50. https://doi.org/10.2307/2033241
Kuang L, Han X, Chen K, Caselli RJ, Reiman EM, Wang Y, I Alzheimer’s Disease Neuroimaging (2019) A concise and persistent feature to study brain restingstate network dynamics: findings from the Alzheimer’s disease neuroimaging Initiative. Hum Brain Mapp 40(4):1062–1081. https://doi.org/10.1002/hbm.24383
Lambiotte R, Rosvall M, Scholtes I (2019) From networks to optimal higherorder models of complex systems. Nat Phys 15(4):313–320. https://doi.org/10.1038/s415670190459y
Lee MH, Smyser CD, Shimony JS (2013) Restingstate fMRI: a review of methods and clinical applications. AJNR Am J Neuroradiol 34(10):1866–1872. https://doi.org/10.3174/ajnr.A3263
Lee Y, Lee J, Oh SM, Lee D, Kahng B (2021) Homological percolation transitions in growing simplicial complexes. https://arxiv.org/abs/2010.12224
Linial N, Peled Y (2016) On the phase transition in random simplicial complexes. Ann Math 184(3):745–773
Liu TT (2016) Noise contributions to the fMRI signal: an overview. Neuroimage 143:141–151. https://doi.org/10.1016/j.neuroimage.2016.09.008
Liu Z, Zhang Y, Yan H, Bai L, Dai R, Wei W, Zhong C, Xue T, Wang H, Feng Y, You Y, Zhang X, Tian J (2012) Altered topological patterns of brain networks in mild cognitive impairment and Alzheimer’s disease: a restingstate fMRI study. Psychiatry Res 202(2):118–125. https://doi.org/10.1016/j.pscychresns.2012.03.002
Maletić S, Rajković M, Vasiljević D (2008) Simplicial complexes of networks and their statistical properties. In: Bubak M, van Albada GD, Dongarra J, Sloot PMA (eds) Computational science ICCS 2008. Springer, Berlin, pp 568–575
Maria C, Boissonnat JD, Glisse M, Yvinec M (2014) The Gudhi Library: Simplicial Complexes and Persistent Homology. Paper presented at the Mathematical Software ICMS 2014, Seoul, South Korea, Aug 5–9
Meunier D, Lambiotte R, Bullmore ET (2010) Modular and hierarchically modular organization of brain networks. Front Neurosci 4:200. https://doi.org/10.3389/fnins.2010.00200
Murphy K, Birn RM, Handwerker DA, Jones TB, Bandettini PA (2009) The impact of global signal regression on resting state correlations: are anticorrelated networks introduced? Neuroimage 44(3):893–905. https://doi.org/10.1016/j.neuroimage.2008.09.036
Najman L, Romon P (2017) Modern approaches to discrete curvature. Lecture notes in mathematics, 1st edn. Springer, Berlin
Newman ME (2008) The mathematics of networks. In: Blume L (ed) The new palgrave encyclopedia of economics, 2nd edn. Palgrave Macmillan, Basingstoke, pp 1–12
Offroy M, Duponchel L (2016) Topological data analysis: a promising big data exploration tool in biology, analytical chemistry and physical chemistry. Anal Chim Acta 910:1–11. https://doi.org/10.1016/j.aca.2015.12.037
Opsahl T, Agneessens F, Skvoretz J (2010) Node centrality in weighted networks: generalizing degree and shortest paths. Social Networks 32(3):245–251. https://doi.org/10.1016/j.socnet.2010.03.006
Otter N, Porter MA, Tillmann U, Grindrod P, Harrington HA (2017) A roadmap for the computation of persistent homology. EPJ Data Science 6(1):17. https://doi.org/10.1140/epjds/s1368801701095
Pardalos PM, Xue J (1994) The maximum clique problem. J Global Optim 4(3):301–328. https://doi.org/10.1007/BF01098364
Petri G, Expert P, Turkheimer F, CarhartHarris R, Nutt D, Hellyer PJ, Vaccarino F (2014) Homological scaffolds of brain functional networks. J R Soc Interface 11(101):20140873. https://doi.org/10.1098/rsif.2014.0873
Phinyomark A, IbáñezMarcelo E, Petri G (2017) Restingstate fMRI functional connectivity: big data preprocessing pipelines and topological data analysis. IEEE Trans Big Data 3(4):415–428. https://doi.org/10.1109/TBDATA.2017.2734883
Plotly Technologies Inc (2015) Collaborative data science. Plotly Technologies Inc, Montréal
Raichle ME (2011) The restless brain. Brain Connect 1(1):3–12. https://doi.org/10.1089/brain.2011.0019
Rosas FE, Mediano PAM, Gastpar M, Jensen HJ (2019) Quantifying highorder interdependencies via multivariate extensions of the mutual information. Phys Rev E 100(3):032305. https://doi.org/10.1103/PhysRevE.100.032305
Rosen BR, Savoy RL (2012) fMRI at 20: has it changed the world? Neuroimage 62(2):1316–1324. https://doi.org/10.1016/j.neuroimage.2012.03.004
Saggar M, Sporns O, GonzalezCastillo J, Bandettini PA, Carlsson G, Glover G, Reiss AL (2018) Towards a new approach to reveal dynamical organization of the brain using topological data analysis. Nat Commun 9(1):1399. https://doi.org/10.1038/s41467018036644
Salch A, Regalski A, Abdallah H, Suryadevara R, Catanzaro MJ, Diwadkar VA (2021) From mathematics to medicine: a practical primer on topological data analysis (TDA) and the development of related analytic tools for the functional discovery of latent structure in fMRI data. PLOS ONE 16(8):e0255859. https://doi.org/10.1371/journal.pone.0255859
Santos FAN, da Silva LCB, CoutinhoFilho MD (2017) Topological approach to microcanonical thermodynamics and phase transition of interacting classical spins. J Stat Mech Theory Exp 1:013202. https://doi.org/10.1088/17425468/2017/1/013202
Santos FAN, Raposo EP, CoutinhoFilho MD, Copelli M, Stam CJ, Douw L (2019) Topological phase transitions in functional brain networks. Phys Rev E 100(3–1):032414. https://doi.org/10.1103/PhysRevE.100.032414
Simpson SL, Bowman FD, Laurienti PJ (2013) Analyzing complex functional brain networks: fusing statistics and network science to understand the brain(*†). Stat Surv 7:1–36. https://doi.org/10.1214/13SS103
Singh G, Memoli F, Ishkhanov T, Sapiro G, Carlsson G, Ringach DL (2008) Topological analysis of population activity in visual cortex. J vis 8(8):11–11. https://doi.org/10.1167/8.8.11
Sizemore AE, Giusti C, Kahn A, Vettel JM, Betzel RF, Bassett DS (2018) Cliques and cavities in the human connectome. J Comput Neurosci 44(1):115–145. https://doi.org/10.1007/s1082701706726
Sizemore AE, PhillipsCremins JE, Ghrist R, Bassett DS (2019) The importance of the whole: topological data analysis for the network neuroscientist. Netw Neurosci 3(3):656–673. https://doi.org/10.1162/netn_a_00073
Smith SM, Vidaurre D, Beckmann CF, Glasser MF, Jenkinson M, Miller KL, Nichols TE, Robinson EC, SalimiKhorshidi G, Woolrich MW, Barch DM, Ugurbil K, Van Essen DC (2013) Functional connectomics from restingstate fMRI. Trends Cogn Sci 17(12):666–682. https://doi.org/10.1016/j.tics.2013.09.016
Smitha KA, Akhil Raja K, Arun KM, Rajesh PG, Thomas B, Kapilamoorthy TR, Kesavadas C (2017) Resting state fMRI: a review on methods in resting state connectivity analysis and resting state networks. Neuroradiol J 30(4):305–317. https://doi.org/10.1177/1971400917697342
Songdechakraiwut T, Chung MK Dynamic Topological Data Analysis for Functional Brain Signals. In: 2020 IEEE 17th International Symposium on Biomedical Imaging Workshops (ISBI Workshops), 4–4 April 2020. p 1–4. doi: https://doi.org/10.1109/ISBIWorkshops50223.2020.9153431
Speidel L, Harrington HA, Chapman SJ, Porter MA (2018) Topological data analysis of continuum percolation with disks. Phys Rev E 98(1):012318. https://doi.org/10.1103/PhysRevE.98.012318
Sporns O (2018) Graph theory methods: applications in brain networks. Dialogues Clin Neurosci 20(2):111–121
Sporns O, Tononi G, Edelman GM (2000) Theoretical neuroanatomy: relating anatomical and functional connectivity in graphs and cortical connection matrices. Cereb Cortex 10(2):127–141. https://doi.org/10.1093/cercor/10.2.127
Stam CJ, Tewarie P, Van Dellen E, van Straaten EC, Hillebrand A, Van Mieghem P (2014) The trees and the forest: characterization of complex brain networks with minimum spanning trees. Int J Psychophysiol 92(3):129–138. https://doi.org/10.1016/j.ijpsycho.2014.04.001
Stolz B (2014) Computational topology in neuroscience. M.Sc. Thesis, University of Oxford
Strother SC (2006) Evaluating fMRI preprocessing pipelines. IEEE Eng Med Biol Mag 25(2):27–41. https://doi.org/10.1109/MEMB.2006.1607667
Suo X, Lei D, Li K, Chen F, Li F, Li L, Huang X, Lui S, Li L, Kemp GJ, Gong Q (2015) Disrupted brain network topology in pediatric posttraumatic stress disorder: a restingstate fMRI study. Hum Brain Mapp 36(9):3677–3686. https://doi.org/10.1002/hbm.22871
van Dellen E, Sommer IE, Bohlken MM, Tewarie P, Draaisma L, Zalesky A, Di Biase M, Brown JA, Douw L, Otte WM, Mandl RCW, Stam CJ (2018) Minimum spanning tree analysis of the human connectome. Hum Brain Mapp 39(6):2455–2471. https://doi.org/10.1002/hbm.24014
van Wijk BCM, Stam CJ, Daffertshofer A (2010) Comparing brain networks of different size and connectivity density using graph theory. PLOS ONE 5(10):e13701–e13701. https://doi.org/10.1371/journal.pone.0013701
van den Heuvel MP, Hulshoff Pol HE (2010) Exploring the brain network: a review on restingstate fMRI functional connectivity. Eur Neuropsychopharmacol 20(8):519–534. https://doi.org/10.1016/j.euroneuro.2010.03.008
van den Heuvel MP, de Lange SC, Zalesky A, Seguin C, Yeo BTT, Schmidt R (2017) Proportional thresholding in restingstate fMRI functional connectivity networks and consequences for patientcontrol connectome studies: issues and recommendations. Neuroimage 152:437–449. https://doi.org/10.1016/j.neuroimage.2017.02.005
Viger F, Latapy M (2005) Efficient and simple generation of random simple connected graphs with prescribed degree sequence. Computing and combinatorics. Springer, Berlin, pp 440–449
Wang J, Wang L, Zang Y, Yang H, Tang H, Gong Q, Chen Z, Zhu C, He Y (2009) Parcellationdependent smallworld brain functional networks: a restingstate fMRI study. Hum Brain Mapp 30(5):1511–1523. https://doi.org/10.1002/hbm.20623
Wang X, Jiao D, Zhang X, Lin X (2017) Altered degree centrality in childhood absence epilepsy: a restingstate fMRI study. J Neurol Sci 373:274–279. https://doi.org/10.1016/j.jns.2016.12.054
Wang Y, Zhao Y, Nie H, Liu C, Chen J (2018) Disrupted brain network efficiency and decreased functional connectivity in multisensory modality regions in male patients with alcohol use disorder. Front Human Neurosci. https://doi.org/10.3389/fnhum.2018.00513
Watts DJ, Strogatz SH (1998) Collective dynamics of ‘smallworld’networks. Nature 393(6684):440. https://doi.org/10.1038/30918
Weber M, Stelzer J, Saucan E, Naitsat A, Lohmann G, Jost J (2017) Curvaturebased methods for brain network analysis. https://arxiv.org/abs/1707.00180. Accessed Feb 2021
Wink AM (2019) Eigenvector centrality dynamics from restingstate fMRI: gender and age differences in healthy subjects. Front Neurosci 13:648. https://doi.org/10.3389/fnins.2019.00648
Wu Z, Menichetti G, Rahmede C, Bianconi G (2015) Emergent complex network geometry. Sci Rep 5:10073. https://doi.org/10.1038/srep10073
Wu Z, Xu D, Potter T, Zhang Y, Alzheimer’s Disease Neuroimaging I (2019) Effects of brain parcellation on the characterization of topological deterioration in Alzheimer’s disease. Front Aging Neurosci 11:113. https://doi.org/10.3389/fnagi.2019.00113
Yu Q, Allen EA, Sui J, Arbabshirani MR, Pearlson G, Calhoun VD (2012) Brain connectivity networks in schizophrenia underlying resting state functional magnetic resonance imaging. Curr Top Med Chem 12(21):2415–2425. https://doi.org/10.2174/156802612805289890
Zalesky A, Fornito A, Bullmore E (2012) On the use of correlation as a measure of network connectivity. Neuroimage 60(4):2096–2106. https://doi.org/10.1016/j.neuroimage.2012.02.001
Zhan L, Jenkins LM, Wolfson OE, GadElkarim JJ, Nocito K, Thompson PM, Ajilore OA, Chung MK, Leow AD (2017) The significance of negative correlations in brain connectivity. J Comp Neurol 525(15):3251–3265. https://doi.org/10.1002/cne.24274
Zomorodian AJ (2005) Topology for computing, vol 16, 1st edn. Cambridge University Press, New York
Acknowledgements
We would like to thank our team members who helped us by testing and providing feedback on the Jupyter Notebooks, Pierre Baudot for sending feedback on the manuscript, NEURASMUS and the 17EURE0028 project for supporting the first author’s studies.
Funding
Not applicable.
Author information
Authors and Affiliations
Contributions
EGZC contributions encompass conceptualisation, software, writing—original draft preparation. GM contributions encompass software, writing—original draft preparation. CV contribution encompasses writing—review & editing. LD contributions encompass conceptualisation, writing—original draft preparation, supervision. FANS contributions encompass conceptualisation, software, writing—original draft preparation, supervision.
Corresponding author
Ethics declarations
Conflict of interest
Not applicable.
Ethics approval
Not applicable.
Consent to participate
Not applicable.
Consent for publication
Not applicable.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Supplementary file1 (MP4 11625 KB) Supplementary Material 1. Video. 3D filtration of networks with 1, 2, 3, and 4cliques
429_2021_2435_MOESM2_ESM.docx
Supplementary file2 (DOCX 47 KB) Supplementary Material 2. Code block. Computation of the Euler characteristic. Supplementary Material 3. Code block. Computation of the Betti numbers. Supplementary Material 4. Code block. Computation of Curvature
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Centeno, E.G.Z., Moreni, G., Vriend, C. et al. A handson tutorial on network and topological neuroscience. Brain Struct Funct 227, 741–762 (2022). https://doi.org/10.1007/s00429021024350
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00429021024350