In lidiaad/GOpro: Find the most characteristic gene ontology terms for groups of human genes

BiocStyle::markdown()

Installation

source("http://www.bioconductor.org/biocLite.R")
biocLite("GOpro", dependencies = TRUE)

Loading

After the package is installed, it can be loaded into R workspace typing

library(GOpro)

Overview

This document presents an overview of the GOpro package. This package is for determining groups of genes and finding characteristic functions for these groups. It allows for interpreting groups of genes by their most characteristic biological function. It provides one function findGO which is based on the set of methods. One of these methods allows for determining significantly different genes between at least two distinct groups (i.e. patients with different medical condition) - the ANOVA test with correction for multiple testing. It also provides two methods for grouping genes. One of them is so-called all pairwise comparisons utilizing Tukey's method. By this method profiles of genes are determined, i.e. in terms of the gene expression genes are grouped according to the differences in the expressions between given cohorts. Another method of grouping is hierarchical clustering. This package provides a method for finding the most characteristic gene ontology terms for anteriorly obtained groups using the one-sided Fisher's test for overrepresentation of the gene ontology term. If genes were grouped by the hierarchical clustering, then the most characteristic function is found for all possible groups (for each node in the dendrogram).

Details

Genes must be named with the gene aliases and they must be arranged in the same order for each cohort.

Determining significantly different genes based on their expressions

Genes which are statistically differently expressed are selected for the further analysis by ANOVA test. The topAOV parameter denotes the maximum number of significantly different genes to be selected. The significance level of ANOVA test is specified by the sig.levelAOV parameter.
This threshold is used as the significance level in the BH correction for multiple testing. In the case of equal p-values of the test (below the given threshold), all genes for which the p-value of the test is the same as for the gene numbered with the topAOV value are included in the result.

Grouping genes based on their similarity

There are two methods provided for grouping genes. They are specified by the grouped parameter. The first one using Tukey's test is called when grouped equals 'tukey' and the second one can be called by using the 'clustering' value.

All pairwise comparisons by Tukey's test

The Tukey's test is applied to group genes based on their profiles. The sig.levelTUK parameter denotes the significance level of Tukey's test. For each gene two-sided Tukey's test is conducted among cohorts. The mean expressions in the cohorts are arranged in ascending order and the result of the test is adapted. All genes with the same order of means and the same result of the test are grouped together. I.e. notation colon\=bladder\<leukemia denotes that the mean expression calculated for a particular gene in the colon cancer cohort is statistically the same as for the bladder cancer cohort. Both mean values determined for aforementioned cohorts are statistically lower than the mean expression measured for the leukemia cohort.

Hierarchical clustering

Hierarchical clustering is utilized for grouping genes based on dissimilarity measures. In this case all clusters are subjected to the further analysis. The clust.metric parameter is a method to calculate a distance measure between genes, the clust.method is the agglomeration method used to cluster genes, and the dist.matrix is an optional parameter for distance matrix if available clust.metric methods are not sufficient for the user.

Finding characteristic gene ontology terms

For each specified group the org.Hs.eg.db is searched for all relevant GO terms. The number of counts of each GO term is calculated for each group. Then the Fisher's test is applied in order to find the most characteristic GO terms for each group of genes. The minGO and maxGO parameters denote the range of counts of genes annotated with each GO term. All GO terms with counts above or below this range are omitted in the analysis. It enables for the exclusion of very rare or very common gene ontology terms. Gene ontology domains to be searched for GO terms can be specified by the onto parameter. Possible domains are: 'MF' (molecular function), 'BP' (biological process), and 'CC' (cellular component). The sig.levelGO parameter specifies the significance level of the Fisher's test (correction for multiple testing is included).

Data

Data used in this example comes from The Cancer Genome Atlas. They were downloaded via RTCGA.PANCAN12 package. The data represents expressions of 300 genes randomly chosen from 16115 genes determined for each patient (sample). Three cohorts are included: acute myleoid leukemia, colon cancer, and bladder cancer. The data is stored in this GOpro package as a MultiAssayExperiment object.

An example of the data structure:

exrtcga

Example

To run the analysis with default parameters on the exrtcga object call:

findGO(exrtcga)

The results of the analysis can be presented in a more descriptive way or in a concise one. To get more descriptive results use extend=TRUE option. Additionally, the TERM, DEFINITION, and ONTOLOGY for each ontology term are returned.

findGO(exrtcga, extend = TRUE)

In order to find top 2 GO terms for genes grouped by hierarchical clustering run the following call. The result of clustering is presented on the plot.

findGO(exrtcga, topGO = 2, grouped = 'clustering')

The plot can be also enriched with information about the most frequent ontology domain for each node on the dendrogram.

findGO(exrtcga, topGO = 2, grouped = 'clustering', over.rep = TRUE)

lidiaad/GOpro documentation built on May 21, 2019, 6:13 a.m.