Winterbach et al. BMC Systems Biology 2013, Topology of molecular interaction networks

Topology of molecular interaction networks

http://www.biomedcentral.com/content/pdf/1752-0509-7-90.pdf

Winterbach et al. BMC Systems Biology 2013, 7:90
http://www.biomedcentral.com/1752-0509/7/90

Algorithms and tools for protein–protein interaction networks clustering, with a special focus on population-based stochastic methods Clara Pizzuti1,† and Simona E. Rombo2,*,†

Algorithms and tools for protein–protein interaction networks clustering, with a special focus on population-based stochastic methods

Clara Pizzuti¹,^† and 

Simona E. Rombo, Bioinformatics 2014.

http://bioinformatics.oxfordjournals.org/content/30/10/1343.short


PR14 used 3 yeast PPI data to compare MCL with others. The MCL parameter was taken from Boheree2006. PR14 used protein complex as 'golden standard'. When overlapping score > 20%, MCL is the best algorithm. Bader's MCODE is also a good method for certain parameter settings.

Wang, Li, Deng, Pan, BMC review on clustering methods for protein interaction networks.

Recent advances in clustering methods for protein interaction networks

Jianxin Wang1,2*, Min Li1*, Youping Deng3, Yi Pan2

From The ISIBM International Joint Conference on Bioinformatics, Systems Biology and Intelligent

Computing (IJCBS), Shanghai, China. 3-8 August 2009

http://www.biomedcentral.com/content/pdf/1471-2164-11-S3-S10.pdf

cited by
http://scholar.google.com/scholar?cites=16432683922097612422&as_sdt=5,43&sciodt=0,43&hl=en

Reviewed 20 clustering methods, including MCL. MCL is commented as the highly successful.

10. Brohée S, van Helden J: Evaluation of clustering algorithms for proteinprotein

interaction networks. BMC Bioinformatics 2006, 7:48.

63. Vlasblom J, Wodak SJ: Markov clustering versus affinity propagation for

the partitioning of protein interaction graphs. BMC Bioinformatics 2009,10:99.

Lin C, Cho Y-R, Hwang W-C, Pei P, and Zhang A. 2007. Clustering Methods in a Protein–Protein Interaction Network. In: Hu X, and Pan Y, eds. Knowledge Discovery in Bioinformatics: John Wiley & Sons, Inc., 319-355.

CLUSTERING METHODS IN PROTEIN-PROTEIN INTERACTION NETWORK
Chuan Lin, Young-rae Cho, Woo-chang Hwang, Pengjun Pei, Aidong Zhang
Department of Computer Science and Engineering
State University of New York at Buffalo

Cite as:
Lin C, Cho Y-R, Hwang W-C, Pei P, and Zhang A. 2007. Clustering Methods in a Protein–Protein Interaction Network. In: Hu X, and Pan Y, eds. Knowledge Discovery in Bioinformatics: John Wiley & Sons, Inc., 319-355.

This review article did not provide enough details on validation and comparison of different algorithms.

MCL algorithm comparison

Brohee and van Helden 2006, BMC [BH06]

BH06 compared Markov Clustering (MCL), Restricted Neighborhood Search Clustering (RNSC), Super Paramagnetic Clustering (SPC), and Molecular Complex Detection (MCODE), using annotated protein complex as bench mark. Random noises were introduced by randomly adding and deleting edges. MCL is the most reliable and robust method. 

disconnected clusters from MCL

Our MCL clusters at high I-values give genes in clusters with disconnected nodes. This is a problem has been discussed before:

http://larsjuhljensen.wordpress.com/2010/03/01/analysis-markov-clustering-and-the-case-of-the-unnatural-clusters/

According to the post, this is not an "error", but a possible though "unexpected" result. Looks like "--force-connected=y" option might solve this problem. I will redo the clustering and then check.