In hyginn/BCB420.2019.ESA: Exploratory Systems Analysis Tools

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

# knitr knits in a new session with an empty global workspace after setting its
# working directory to ./vignettes. To make your package functions available in
# the vignette, you have to load the library. The following two lines should
# accomplish this without manual intervention:
pkgName <- trimws(gsub("^Package:", "", readLines("../DESCRIPTION")[1]))
library(pkgName, character.only = TRUE)

There are many links and references in this document. If you find anything here ambiguous, inaccurate, outdated, incomplete, or broken, please [file an issue](https://github.com/hyginn/BCB420.2019.ESA/issues)!

About this Vignette

This Vignette explains the findSTRINGKnodes() function that is part of the BCB420.2019.ESA package. The function finds the strength of association (Knode value) of each gene in the p>=0.9 STRING functional protein association network. The Knodes metric is based on a modified version of Ripley's K-statistic, as implemented by @Cornish2014.

Function

Usage

findSTRINGKnodes(sys)

where sys is a 5-character system code corresponding to a system collected in "SysDB". The function returns a list of K node values for each HGNC symbol found in the STRING edges, sorted in descending order. Additionally, the function plots a labelled ggplot2 scatterplot with the top \emph{n} genes, where \emph{n} is the number of genes in sys to visualize the genes that did have a high association to the system.

K node calculation

The findSTRINGKnodes() function uses the network of protein associations at a confidence of p >= 0.9 curated by @steipe-STRING to calculate strength of association to a given system of genes available through the "SysDB" dataset. A graph is constructed from the STRING edges, and the vertex representing genes which are in the system have a weight of 1. All other vertices have a weight of 0.

Then, the modified Ripley's K-statistic is calculated through:

$$K^{node_i[s]} = 2/p * \sum_j(p_j - \bar{p}) (dg(i,j)<=s)$$

where $p_j$ is the weight of vertex $j$, $\bar{p}$ is the mean vertex weight across all vertices, and $I(dg[i,j]<=s)$ is an identity function equaling 1 if vertex i and vertex j are within distance s and 0 otherwise, as described by @Cornish2014.

Visualization

The resulting Knode values are plotted in a labelled scatterplot using ggplot2. Only the number of genes equal to the size of the input system is plotted. Genes that are part of the input system are labelled in green, and those that are not are labelled in red.