HiVis: a portable, scalable tool for hierarchical visualization and analysis of biological networks

We use a top-down approach to cluster the biological networks into a multi-level node-link graph. We demonstrate the main features in Fig. 1 through a data set from STRING interaction database (https://string-db.org/cgi/input.pl). It is well known that many biological networks contain clique and hub genes which are highly clustered and contain self-similar structure. Therefore, clustering them together into one node can reveal their hierarchical information and make the visualization readable.

Currently, we provide two cluster algorithms, k-means clustering, and fast approximate spectral clustering (Yan et al. 2009), which is faster than normal spectral clustering method and is more suitable for network clustering than k-means algorithm. The idea is to use k-means as a preprocessor and then to perform the normal spectral clustering method. Here, we also implemented a heuristic approach to determine the number of clusters by checking the slope of neighboring points of sorted eigenvalues.

It is difficult to display the biological network meaningfully in one scope, HiVis solves it by cluster them into a hierarchical tree and only reveal the nodes in the current view and the view in the next hierarchy. To be more specific, here, we present an example illustrated in Fig. 1, 1a is the overview of HiVis, the genes are clustered using k-means, and each node represents a group of them. It can switch the algorithm to modified spectral clustering and the result is shown in Fig. 1d. HiVis offers smooth animation methods to further analyze the computed clusters. When selecting one cluster, it will show a full scope of its inner structure; as shown in Fig. 1b, users can also return to the previous stage, or they can further analyze the current cluster with the same procedure. Apart from that HiVis will show each genes within the cluster, as is shown in the left part of Fig. 1b, when choosing one gene, HiVis will display the gene and its most related genes in a network, as is shown in Fig. 1c. Thus, it will help to analyze the network locally.

In this paper, we present HiVis, a software built on electron for hierarchical visualization and analysis of biological networks. We implemented the k-means clustering, fast approximate spectral clustering algorithm, and force-direct layout algorithm to let HiVis capable of handling large-scale biological data. It is worth noting that this kind of displaying technique is very general and can be applied to other networks like social network, security network, etc.