Discussion

Understanding the community of proteins in a protein protein network is important to find the functionality and the relevance of a specific macro-molecular assemblies or even in discovering proteins affecting a specific biological process. Effective algorithms can enhance drug discovery and diseases treatments.

In this work, I leveraged the idea of hypergraph partitioning to detect communities in a protein protein network. I used cross validation to evaluate the quality of the detected partitions. I was able to predict 33 strong communities in the Saccharomyces cerevisiae protein-protein interaction network. The evaluation results show that there is a dense relationship between proteins in these communities such that removing 50% of the edges in the network does not break these communities. I observed that the hypergraph partitioner is good at predicting larger sized communities while it does not generate good quality small sized communities. In general, predicting an edge between two protein requires more samples and experiments than predicting a community of proteins. I have not studied these 33 communities separately to be able to guess about the structure of the community but this would be an interesting study to conduct to further understand these communities. Another interesting future work can be the study of balance criteria. The quality of detected communities can be improved by changing the balance criteria in the hypergraph partitioner.