GSAn

Documentation


GSAn is a new alternative to gene set analysis exploiting the semantic of concepts in the Gene Ontology (GO). Combining data mining and combinatorial algorithms, GSAn provides a reduced and synthetic number of GO terms resuming the biological role of the gene set input. In the 1st version, GSAn includes the three ontologies of GO: biological process (BP), molecular function (MF) and cellular component (CC). The sections of interest are:

Summary

1. Start Analysis

2. Visualization

3. Results

4. Formulas

1. Start Analysis


NOTE: If the organism is uploaded, users have to verify that the proteins are in the first column and the genes in the second column. In that case, you can use other ids matching with the proteins or genes.

Advanced parameters


2. Visualization


3. Results



Understanding the gauge figures


Understanding the synthetic annotation


Bubble chart

We display results according to a visualization combining two classes: node-link and space-filling visualization. These visualizations are dynamic and users can zoom to get a deeper information. The complete information about this visualization can be found in Ayllon-Benitez et al.



4. Formulas

Information content

The information content, or IC, is a score associated with the terms, explaining how much they are informative. The bigger the IC is, the more specific the term is. Two kinds of IC are used in GSAn: intrinsic and extrinsic. The intrinsic IC uses only information available within the ontology structure and the extrinsic IC uses external information. In most cases, the IC of a term is computed from the probability of the term as follows:

Resnik or extrinsic

In GSAn, only one kind of extrinsic IC is used. The used IC was proposed by Resnik in 1995. This IC uses the frequency of occurrence of a word in a given document. In the Gene Ontology, the frequency of occurrence (number of genes) is computed according to the GOA file of each organism. This IC is applied with some semantic similarity measures (see formulas below). Thus, the extrinsic IC is computed as follows:


Note: a term includes all the genes and proteins associated with it and its descendants using the true path. The true path means that the information provided by a term is inherited by its descendants, thus if a gene is associated with a term that is descendant of another term, then this gene is also associated with the ancestor term.

Mazandu

The intrinsic IC used in GSAn was proposed by Mazandu and Mulder taking into account the position of the terms in the graph. First, the probability is computed exponentially, from the root term to the leaf terms. Each term depends on the probability of its parents divided by the children of the parent. Thus, the IC provided by Mazandu for a term t is the minus logarithm of the probability.


Semantic Similarity

Resnik

That semantic similarity measure was proposed by Resnik at the same time than the IC, where the similarity between two terms corresponds to the IC of their most informative common ancestors (or currently called the MICA). The equation is as follow and we normalized the measures by dividing them by the IC max:


Lin

To improve the results of Resnik's similarity, Lin proposed another normalization by dividing twice the IC of MICA by the addition of IC between the compared terms.


AIC

The aggregate information content or AIC is an alternative measure proposed by Song et al. combining the semantic value of Wang (not used here) and the Resnik IC. The semantic value and the semantic weight are computed as follows:


The semantic similarity is according to the similarity provided by Wang et al., but using the SV and SW shown above as follows:

NUnivers

Using the IC proposed by Mazandu and Mulder., the similarity of NUnivers between two terms computes the most informative content ancestor divided by the maximal IC between the compared terms.


Distance Function

The distance function (DF) (used by Quesada-Martinez et al.) combines the edge semantic similarity and node based. The measure uses the Jaccard similarity of the union and intersection of the ancestor between the terms to compare.

Gene set similarity

GS²

This gene set similarity computes a similarity by taking into account the annotation terms and their ancestors of all genes involved in a particular gene set. The formula that we applied is as follows (Ruths et al.):



where G is the gene set, TGO(x) is the list of GO terms associated with a given gene x, ancestors(y) is the list of ancestor terms of the GO term y in addition to y itself, rankG−x(z) is the number of genes in the set except for the gene x that is associated with the GO term z. When the value is 0, all genes have distinct annotation, and when the value is 1, all genes have the same annotation. A low similarity value (under 0.5) can result in a unprecise GSAn annotation.