Beyond gene data analysis: A new data mining approach for interactive and correlative investigation of molecular and clinical data

Abstract

576

Objectives: The currently available software used for gene data analysis focused on the evaluation of gene-gene data. This study proposes a new data mining approach for investigating the interaction not only of the gene-gene data but also gene-PET (molecular-clinical) data by applying correlation and regression analysis as well as discriminant analysis. The implemented software can be used to improve the diagnostics and therapy management and identify possible targets for new PET radiopharmaceuticals.

Methods: The software is integrated to the PMod software (PMOD Technologies Ltd., Adlisvil, Switzerland). The Excel data format is selected as the input standard of gene chip data and PET data. Based on the Java implementation, the software is tested and evaluated on different platforms. Dynamical PET data were obtained from patients with giant cell tumors and colorectal tumors, undergoing PET FDG examinations. Furthermore, tumor specimen were obtained by surgery and evaluated with gene chip technology (Affymetrix U95A and U123A).

Results: Based on the Excel input format, the data can be easily transformed to other standard formats within the PMod software. Using the powerful image tool of PMod, all gene expression values can be directly displayed as an image. The functions of gene-gene analysis module include selection of feature genes (e.g. all genes with the term “receptor” in its description), comparison of gene values of different patient groups (e.g. Tumor VS Normal), 2D/3D gene array plot graphs and so on. The software provides a lot of enhanced gene annotation information though linking each gene to external web pages like Affymetrix and PubMed. After loading the gene data file and PET file, gene data and PET parameters are displayed and changed synchronously according to diagnostic cases and patients. The calculated correlation coefficients of corresponding gene-PET vectors are also displayed as an image and can be adjusted at different significant probability. Furthermore, gene-PET vectors can be analyzed with comprehensive regression functions that provide many optional choices such as gene probe code, patient classification, PET parameters, regression models and probability of significance.

Conclusions: Gene data and quantitative PET data are no longer isolated from each other. The new data mining approach provided a very unique tool for the combined analysis of gene and PET data. The platform can be used to investigate the relationship between any two sets of data such as large-scale gene data and other clinical data.