学术文献库

The conventional approach for analyzing gene expression data involves clustering algorithms. Cluster analyses provide partitioning of the set of genes that can predict biological classification based on its similarity in n-dimensional space. In this study, we investigate whether network analysis will provide an advantage over the traditional approach. We identify the advantages and disadvantages of using the value-based and the rank-based construction in creating a graph representation of the original gene-expression data in a time-series format. We tested four community detection algorithms, namely, the Clauset-Newman-Moore (greedy), Louvain, Leiden, and Girvan-Newman algorithms in predicting the 5 functional groups of genes. We used the Adjusted Rand Index to assess the quality of the predicted communities with respect to the biological classifications. We showed that Girvan-Newman outperforms the 3 modularity-based algorithms in both value-based and ranked-based constructed graphs. Moreover, we also show that when compared to the conventional clustering algorithms such as K-means, Spectral, Birch, and Agglomerative algorithms, we obtained a higher ARI with Girvan-Newman. This study also provides a tool for graph construction, visualization, and community detection for further analysis of gene expression data.