11
Node perturbation study of Protein-Protein Interaction Network VIJEESH T M120443CS GUIDED By: Jayaraj P B

Node Perturbation Study of Protein-Protein Interaction Network

Embed Size (px)

Citation preview

Page 1: Node Perturbation Study of Protein-Protein Interaction Network

Node perturbation study of Protein-Protein Interaction

Network VIJEESH T

M120443CSGUIDED By: Jayaraj P B

Page 2: Node Perturbation Study of Protein-Protein Interaction Network

Problem Statement

Device a parallel method to calculate centrality measures of proteins in a

protein interaction network and implement on GPU. Study the effect of

removing a protein from a protein interaction network on these measure.

Page 3: Node Perturbation Study of Protein-Protein Interaction Network

Centrality measures• Identify the importance of a node in a graph

• E.g. Degree, closeness and betweenness

• Centrality measures can be used to rank proteins according to their essentiality. This helps to identify lethality proteins and drug targets. [1]

Inverse of the sum of distance to all other nodes – closeness[3]

Fraction of shortest paths between other nodes –betweenness[2]

Page 4: Node Perturbation Study of Protein-Protein Interaction Network

Plan of Actions

Identify a parallel algorithm to compute All-Pair-Shortest Path on Graph Develop a parallel method to compute betweenness centrality Identify sparse graph storage representation Implement on GPU and compare performance with CPU

• Identify a PIN corresponding to a particular disease

• Find the relationship between important proteins and disease causing proteins.

Page 5: Node Perturbation Study of Protein-Protein Interaction Network

Work done before last evaluation

• Identified parallel algorithm to compute All Pair Shortest Path.• Delta-stepping algorithm [5]

• Identified faster algorithm to compute betweenness centrality[4]

• Modified delta-stepping algorithm to compute betweenness centrality

• Implemented Delta-stepping algorithm on GPU

• Identified sparse matrix storage representation –CSR [6]

Page 6: Node Perturbation Study of Protein-Protein Interaction Network

Work done after last evaluation

• Implemented betweenness centrality on GPU

Challenges• Storage of predecessors information.

• Lack of stack data structure.

Page 7: Node Perturbation Study of Protein-Protein Interaction Network

Faster algorithm for BC

• Instead of computing pair-dependency, compute the dependency of vertex s on a vertex v

• Brandes[4] algorithm give recursive equation for dependency dep(u) = *(1 + dep(v))

Page 8: Node Perturbation Study of Protein-Protein Interaction Network

The Modified algorithm

• For each source in V• Scan the edges for relaxations

• During relaxation, update predecessor, sigma and level

• For each edge in parallel update the dependency [4]

dep(u) = dep(u) + *(1 + dep(v))

• For each vertex in parallel update the centrality in bottom-up manner

Page 9: Node Perturbation Study of Protein-Protein Interaction Network

Results• Implemented betweenness centrality and closeness centrality on GPU

• Will work for both weighted and unweighted graphs.

facebook yeast PIN0

100

200

300

400

500

600

CPU GPU

Page 10: Node Perturbation Study of Protein-Protein Interaction Network

Future work

• Identification of PIN data

• Include other metric like influential spreading

• Identify the relation between centrality values and disease causing proteins.

Page 11: Node Perturbation Study of Protein-Protein Interaction Network

References[1] H. Jeong, S. P. Mason, A. L. Barabasi, and Z. N. Oltvai, “Lethality and centrality in protein

networks," Nature, vol. 411, no. 6833, pp.41-42, May 2001.

[2] L. C. Freeman, “A set of measures of centrality based on betweenness," Sociometry, vol. 40, no. 1, pp. 35-41, March 1977.

[3] G. Sabidussi, “The centrality index of a graph," Psychometrika, vol. 31, no. 4, pp. 581-603, 1966.

[4] U. Brandes, “A faster algorithm for betweenness centrality," Journal of Mathematical Sociology, vol. 25, pp. 163-177, 2001.

[5] U. Meyer and P. Sanders, “Delta-stepping: a parallelizable shortest path algorithm," Journal of Algorithms, vol. 49, no. 1, pp. 114 -152, 2003

[6] N. Bell and M. Garland, “Efficient sparse matrix-vector multiplication on CUDA," NVIDIA Corporation, NVIDIA Technical Report NVR-2008-004, Dec. 2008.