For neighbourhood based metrics the methods which are implemented are (i) common neighbours (ii) jaccard coefficient\cite{bib16} (iii) cosine similarity (iv) hub promoted index\cite{bib17} (v) hub depressed index (vi) Adamic Adar index \cite{bib18} (vii) Preferential attachment\cite{bib19} (viii) Resource allocation\cite{bib20} (ix) Leicht-Holme-Nerman Index \cite{bib21}. Similarly using path-based metrics one can compute paths between two nodes as similarity between node pairs. The methods are:
- The local path based metric \cite{bib22} uses the path of length 2 and length 3. The metric uses the information of the nearest neighbours and it also uses the information from the nodes within length of 3 distances from the current node.
- The Katz metric\cite{bib23} is based on similarity of all the paths in a graph.This method counts all the paths between given pair of nodes with shorter paths counting more heavily. Parameters are exponential.
- Geodesic similarity metric calculates similarityscore for vertices based on the shortest paths between two given vertices.
- Hitting time\cite{bib24} is calculated based on a random walk starts at a node x and iteratively moves to a neighbor of x chosen uniformly at random. The Hitting time $H_{x,y}$ from x to y is the expected number of steps required for a random walk starting at x to reach y.
- Random walk with restart\cite{bib8,bib24,bib25} is based on pagerank algorithm \cite{bib26}. To compute proximity score between two vertexes we start a random walker at each time step with the probability 1 - c, the walker walks to one of the neighbors and with probability c , the walker goes back to start node. After many time steps the probability of finding the random walker at a node converges to the steady-state probability.
The significance of interaction of links is based on random permutation testing. A random permutation test compares the value of the test statistic predicted data value to the distribution of test statistics when the data are permuted. Supporting Information S1\_NetpredictorVignette provides tutorial for this netpredictor standalone R package.In the web application app one can load their own data or can use the given sample datasets used in the software. For the custom dataset option one needs to upload bipartite adjacency matrix along with the drug similarity matrix and protein sequence matrix. From the given datasets Enzyme, GPCR, Ion Channel and Nuclear Receptor in the application one can load the data and set the parameters for the given algorithms and start computations. The data structure the web application accepts matrix format files for computation. A summary of the contents of each of the tabs shiny netpredictor application is reported in Table 2.