Navigating gene networks

Description

In August and September of 2007, I did a summer job at the electrical engineering department of the K.U.Leuven. The subject was 'Navigating gene networks by text mining and statistical analysis'. The project was mainly divided into two parts, the text mining and the statistical analysis.

In the text mining part, our goal was to create a gene network based on research papers, visualizing which human genes influence each other in some way. So a cocitation of two genes in the same medical publication is considered as a link between these two genes. Ultimately, a file was created containing all the gene links. This gene network was imported into a program called BioLayout, to produce 2D and 3D gene network visualizations like the one below.

A 3D gene network in BioLayout

The statistical part was about studying the differences between human genes that are known to cause genetic disorders when there is something wrong with them ('disease genes') and other, regular genes. A number of properties of the genes were tested, among others the total gene length, the number and length of the parts of the gene that are expressed (e.g. code for proteins - called exons) and the homology of the genes in different species (e.g. if the gene is present in a specific animal this represents the similarity of the animal gene and the human one). The differences between disease and non-disease genes were visualized using boxplots, histograms and pie charts.

At the end, both parts of the project were combined. The gene network was searched for first neighbours of disease genes. Then these neighbours were compared to the disease and non-disease genes on some of the same properties as in the statistical analysis part. We alo checked whether disease genes have a larger number of disease neighbours, or neighbours in general, than non-disease genes.

The ultimate goal of all this is to provide concrete info on how to distinguish between disease and non-disease genes, e.g. create some 'probability value' for each gene to be involved in genetic disorder. An example of this can also be found on the Prospectr website.

People

Many thanks to Bart De Moor and the bioinformatics research group at the K.U.Leuven.