Rosy Sarmah

Weighted edge based clustering to identify protein complexes in proteinprotein interaction networks incorporating gene expression profile

Protein complex detection from protein-protein interaction (PPI) network has received a lot of focus in recent years. A number of methods identify protein complexes as dense sub-graphs using network information while several other methods detect protein complexes based on topological information. While the methods based on identifying dense sub-graphs are more effective in identifying protein complexes, not all protein complexes have high density. Moreover, existing methods focus more on static PPI networks and usually overlook the dynamic nature of protein complexes. Here, we propose a new method, Weighted Edge based Clustering (WEC), to identify protein complexes based on the weight of the edge between two interacting proteins, where the weight is defined by the edge clustering coefficient and the gene expression correlation between the interacting proteins. Our WEC method is capable of detecting highly inter-connected and co-expressed protein complexes. The experimental results of WEC on three real life data shows that our method can detect protein complexes effectively in comparison with other highly cited existing methods. AVAILABILITY The WEC tool is available at http://agnigarh.tezu.ernet.in/~rosy8/shared.html.

An effective graph-based clustering technique to identify coherent patterns from gene expression data

This paper presents an effective parameter-less graph based clustering technique (GCEPD). GCEPD produces highly coherent clusters in terms of various cluster validity measures. The technique finds highly coherent patterns containing genes with high biological relevance. Experiments with real life datasets establish that the method produces clusters that are significantly better than other similar algorithms in terms of various quality measures.

An effective density-based hierarchical clustering technique to identify coherent patterns from gene expression data

We present an effective tree-based clustering technique (Gene ClusTree) for finding clusters over gene expression data. GeneClusTree attempts to find all the clusters over subspaces using a tree-based density approach by scanning the whole database in minimum possible scans and is free from the restrictions of using a normal proximity measure [1]. Effectiveness of GeneClusTree is established in terms of well known z-score measure and p-value over several real-life datasets. The p-value analysis shows that our technique is capable in detecting biologically relevant clusters from gene expression data.

A modified QT-clustering algorithm over Gene Expression data

Clustering is often one of the first steps in Gene Expression Analysis. In this paper we propose a modified-QT clustering algorithm for gene expression datasets that uses a modified Pearsons correlation measure to identify the clusters in gene expression data. Experimental results show the efficiency of the proposed method over several real-life datasets. The proposed method has been found to be better than other comparable algorithms in terms of z-score and p-value measures of cluster quality.

CNNC: a common nearest neighbour clustering approach for gene expression data

We present an effective common nearest neighbour-based clustering technique (CNNC) for finding clusters over gene expression data. CNNC attempts to find all the clusters over gene expression data qualitatively. Our algorithm works by finding clusters using a nearest neighbour-based approach. A regulation-based module for finding sub clusters is also presented here. CNNC was tested on several real-life datasets and the effectiveness is established in terms of well known z-score measure and p-value over several real-life datasets. Using z-score analysis we show that CNNC outperforms other comparable algorithms. The p-value analysis shows that our technique is capable in detecting biologically relevant clusters from gene expression data.

Variable density spatial data clustering

This paper presents an effective clustering method which can detect embedded and nested clusters over variable density space. The proposed method, VDSC uses a density based approach for detecting clusters of arbitrary shapes, sizes and densities. VDSC was compared with several other comparable algorithms and the experimental results show that our method could detect all clusters effectively.

A Density-Based Clustering for Gene Expression Data Using Gene Ontology

Gene expression clustering is built on the premise that similarly expressed genes are included in the same kind of biological process. Recent research has focused on the fact that incorporation of biological knowledge such as gene ontology (GO) improves the result of clustering. This paper demonstrates a Semi-supervised Density-based Clustering (SDC) which uses GO to detect positive and negative co-regulated patterns from the noisy gene expression data. SDC improves a previous algorithm DenGeneClus (DGC) which could handle only positive co-regulation and did not include GO in the clustering process. Experimental results on four real-life data show that SDC outperforms DGC based on z-score and gene ontology enrichment analysis.

A common neighbor based technique to detect protein complexes in PPI networks

Detection of protein complexes by analyzing and understanding PPI networks is an important task and critical to all aspects of cell biology. We present a technique called PROtein COmplex DEtection based on common neighborhood (PROCODE) that considers the inherent organization of protein complexes as well as the regions with heavy interactions in PPI networks to detect protein complexes. Initially, the core of the protein complexes is detected based on the neighborhood of PPI network. Then a merging strategy based on density is used to attach proteins and protein complexes to the core-protein complexes to form biologically meaningful structures. The predicted protein complexes of PROCODE was evaluated and analyzed using four PPI network datasets out of which three were from budding yeast and one from human. Our proposed technique is compared with some of the existing techniques using standard benchmark complexes and PROCODE was found to match very well with actual protein complexes in the benchmark data. The detected complexes were at par with existing biological evidence and knowledge.

Identification of protein complexes in protein-protein interaction networks by core-attachment approach incorporating gene expression profile

Due to the advancement in Proteomic technologies, bulk data of protein-protein interactions (PPI) are available which give researchers in bioinformatics the opportunity to explore and understand biological properties and structure from a networking perspective. Identification of protein complexes is a challenge that has emerged as an attraction to researchers particularly in computational biology. Various computational approaches were developed to identify protein complexes in PPI networks. In this paper, we give a new method based on the core-attachment approach with incorporation of gene expression data known as core-attachment with gene (CAG) expression to identify protein complexes in PPI networks. Experiment results support that our method CAG can detect protein complexes effectively. Validation by biological information, namely co-localisation and gene ontology semantic similarity score reveals that the complexes predicted by our method has high biological relevance. We also give a comparison of our method with four other popular methods in the field.

A Fuzzy Graph Based Cluster Affinity Search Technique for clustering of gene expression data

Cluster analysis is a widely used data mining technique for extracting biological knowledge from gene expression data. In this paper, we modified one of the graph-theoretic approach CAST by using fuzzy graph concept. Our algorithm FGBCAST (Fuzzy Graph Based Cluster Affinity Search Technique) is tested over three real life datasets Yeast Cell Cycle, Yeast Sporulation and Escheria Coli. The performance of the proposed algorithm gives better results than CAST in terms of z-score, p-value and Q-value.