244

Bioinformatics of the Brain

9.8

Conclusions

We have provided a graph-theoretical analysis of complex brain networks in

this chapter. The brain network needs to be constructed using the neuroimag-

ing process and then processing obtained data to yield the connectivity matrix

first. The adjacency matrix is then formed by filtering the connectivity ma-

trix over which the graph representing the connectivity of the brain nodes is

constructed. This graph has the small-world and scale-free properties as var-

ious other complex biological networks such as protein interaction networks

and metabolic pathways. We then reviewed main analysis parameters that

can be used for the analysis of brain networks which are clustering coefficient,

matching index, and centrality. A brain network has densely connected re-

gions called modules or clusters and within each cluster, hubs are the nodes

with very high number of connections to their neighbors. We provided basic

module search algorithms that can be applied to brain networks with modu-

larity maximization oriented algorithms commonly used for this purpose. A

repeating subgraph of a larger graph is called a motif and brain networks have

motifs which may represent some basic function performed by them. Discov-

ery of motifs is a computationally hard problem and various heuristics are

used to find them. We reviewed motif discovery in brain networks and various

methods to evaluate the significance of the motifs found in brain networks.

[54].

Network alignment aims to find similarity between two or more networks.

Finding similarity between a brain network with a disease condition such as

AD, ADHD, MS or autism and a suspected patient may guide health profes-

sionals to diagnose these diseases. We also surveyed the brain network struc-

ture alterations in neurological disorders with emphasis on AD, schizophrenia

and PD. General findings in various studies outlined were reduced or ab-

normal connectivity of hubs which connect brain functional regions, reduced

clustering coefficient and increased characteristic path length causing the loss

of small-world and scale-free features of brain networks, thus implying the

cause of the disorder.

Our general conclusion is that representing brain activities as a graph

provides various graph algorithms, methods and tools to be readily available

for the analysis of brain networks to aid our understanding of disease states

and the cognitive processes of the brain. Research studies in this area, namely

graph-theoretic analysis of brain networks, may lead to improved diagnosis

and treatments of neurological disorders resulting in clinical outcomes.