A Web-based platform to find out relations between OMIC data and clinical features

Abstract

We developed a platform that visualizes all the dimensions of so-called ‘OMIC’ data from genomic, transcriptomic, and proteomic domains, and helps users identify interesting data dimensions that might be associated with a set of clinical features of diseases. For this, we organized the textual descriptions of Clinical Synopsis fields in OMIM (Online Mendelian Inheritance in Man) into a quantitative format, and developed a Web-based interactive and graphical search system.

Problems addressed

Analysis of OMIC data may reveal underlying genetic interactions and lead to the understanding of disease processes. One of the important analyses would be to find out shared OMIC features among the genes associated with diseases manifesting similar clinical features. The first problem would be to measure the similarity between two sets of clinical features that are usually described in a variety of medical terms. The second problem would be to deal with the multi-dimensionality of both OMIC data and clinical features.

Purpose

First, we aimed to map the clinical features of diseases to relevant tissues or biological processes, so that the clinical features could be described with a relatively small number of items that would allow comparison of the clinical features between diseases. Second, we aimed to develop a tool that would visualize all the data points and enable us to find out the correlation between dimensions of OMIC data and clinical features of the diseases.

Methods

To obtain the links between the genes and the clinical features of the diseases, we used OMIM [1] Clinical Synopsis (CS). We selected subheadings of CS that represent affected body parts, tissues, or systems as well as the resultant pathophysiology, e.g. neoplasm. We merged the subheadings that would be linked to the same body parts or systems.

We modeled each disease as an array of cells representing a CS feature filled with the number of descriptions for that CS feature. We hierarchically clustered the disease vectors based on the average Euclidian distance between groups, using Cluster [2] software.

The OMIC data we used was a set of genome-wide gene expression patterns that measured the relative concentration of transcripts in 50 types of normal human tissues using a competitive PCR-based method. We clustered the gene expression patterns in the same way as clinical features.

We ordered the clustered genes and diseases on the X and Y axes, respectively, plotted the genes that had both the gene expression pattern and associated disease on the X-Y plain, and stored the coordinates. We made a system consisting of a server database and a graphical user interface implemented as a Macromedia Flash plug-in on Web browsers. The system enabled interactive searches of the genes on selected areas on the plain.

Results and Discussion

We selected 169 out of to 201 subheadings in CS, and organized them into 51 categories. These categories accounted for 41,416 out of 56,589 descriptions, or 73% of the descriptions in CS.

The database with a Web-based graphical user interface, named Gene/OMIM Viewer, is accessible at http://www.genelexpo.jp/GeneOMIMViewer/index.html. The patterns of gene expression and disease clinical features are shown as heat maps on the sides of X and Y axes, respectively. Select a rectangular area on the X-Y plain, and the relative density of the plotted genes in this area compared to the surrounding areas, and the gene list with links to OMIM will be shown. Options to display all the genes or diseases that make the edges of this area would help users narrow down the space in which to search for relevant diseases or genes.