Example

For the example a graph is generated by spike_in().

By default it generates a table with 500 genes. These 500 genes are bimodal genes which list 500 patients with a 70:30 proportion into low and high expressing patients. These groups are called modalities. The modalities are connected by their patient composition. This composition is completely randomized by the function sample() out of the 500 patients. The distance between the means of the two modalities is set to 2, which is the fold change for this gene. These genes are called noise.

In this noise a spike is insert. By default it also contains 500 patients with a 70:30 proportion and a FC of 2. But only 25 genes are generated for the spike. The first part of the patient composition in the spike is performed in the same way as in the noise. To this randomized connection a overlap between the genes of the noise is added. By default this value is set to 65%, which means that 65% of the patients in one modality of the spike can be found in an other modality of the spike.

library(PARrOT)
table <- spike_in()

The generated table looks like:

This table is then given to the calcscorematrix() by:

matrix <- calcscorematrix(table = table)

This generates a histogram for all weights before and after the normalization and a PDF which contains a heatmap before and after normalization. The generated adjacency matrix is structured as follows:

As mentioned in the readme the amount of 500 genes or rather 1000 modalities and 1000^2 connections (edges) would be to big workload to compute in an acceptable amount of time. For this reason the modalities are here reduced to genes and only the top 20 (by default) edges per gene are taken. This reduction is made by:

reduced_graph <- buildSinglenode(matrix = matrix)

This command generates a histogram from all edges after the reduction and the adjMatrix_singlenode.csv file which contains 3 rows where the first two contain a vertex and the third weight of the edge between them, which looks like this: