Summary: Efficient distributed algorithms for pattern detection in graphs derived from
Guillermo A. Cecchi1, A. Ravi Rao1, Dante R. Chialvo2, Avi Ma'ayan3, Maria V. Centeno2 and A. Vania Apkarian2
1IBM T. J. Watson Research Center, Yorktown Heights, NY 10598, USA
2Dept. of Physiology, Feinberg School of Medicine, Northwestern University, Chicago, IL 60611
3Dept. of Pharmacology and Biological Chemistry, Mount Sinai School of Medicine, New York, NY 10029, USA.
1p 2p 3p
Recently, we introduced an approach to study topological properties of functional brain
networks1. We demonstrated that the graphs determined by the structure of pairwise
correlations between voxels display very robust topological statistical regularities, including
power-law connectivity scaling and small-worldness, that are shared among other large-scale
biological and technological networks.
One difficulty in extending this analysis, and applying it to obtain a useful discriminatory power
between different brain states, is that the computations become intractable very easily as one
moves up from two-point correlations. Here we present a novel approach that extends our
previous findings to include directional links, and based on this analyze the presence and
significance of higher-order correlation patterns.
In order to tackle the inherently costly computational demands, we developed a series of
algorithms implemented on distributed platforms that render our approach feasible.
In our previous study, we defined graphs embedded in the functional networks by thresholding