Gaussian Graphical Modeling Reveals Specific Lipid Correlations In Glioblastoma Cells

Nikola S. Mueller,Jan Krumsiek,Fabian J. Theis,Christian Boehm,Anke Meyer-Baese

DOI: https://doi.org/10.1117/12.884196

2011-01-01

Abstract:Advances in high-throughput measurements of biological specimens necessitate the development of biologically driven computational techniques. To understand the molecular level of many human diseases, such as cancer, lipid quantifications have been shown to offer an excellent opportunity to reveal disease-specific regulations. The data analysis of the cell lipidome, however, remains a challenging task and cannot be accomplished solely based on intuitive reasoning. We have developed a method to identify a lipid correlation network which is entirely disease-specific. A powerful method to correlate experimentally measured lipid levels across the various samples is a Gaussian Graphical Model (GGM), which is based on partial correlation coefficients. In contrast to regular Pearson correlations, partial correlations aim to identify only direct correlations while eliminating indirect associations. Conventional GGM calculations on the entire dataset can, however, not provide information on whether a correlation is truly disease-specific with respect to the disease samples and not a correlation of control samples. Thus, we implemented a novel differential GGM approach unraveling only the disease-specific correlations, and applied it to the lipidome of immortal Glioblastoma tumor cells. A large set of lipid species were measured by mass spectrometry in order to evaluate lipid remodeling as a result to a combination of perturbation of cells inducing programmed cell death, while the other perturbations served solely as biological controls. With the differential GGM, we were able to reveal Glioblastoma-specific lipid correlations to advance biomedical research on novel gene therapies.