PP.28.24: INTEGRATIVE DATA ANALYSIS IN THE INGENIOUS HYPERCARE COHORT. (June 2015)

• Record Type: Journal Article
• Authors: Robinson, S.; Perco, P.; Husi, H.; Köck, T.; Mischak, H.; Jennings, A.; Monleon, D.; Ravassa, S.; Leenders, J.; Jankowski, J.; Chatzikyrkou, C.; Pieske, B.; Staessen, J.A.; Padmanabhan, S.; Zanchetti, A.; Dominiczak, A.F.; Delles, C.; Consortium, Eu-Mascara
• Is Part Of: Journal of hypertension. Volume 33(2015)Supplement 1
• Publication Date: 2015-06
• Subjects: Hypertension -- Periodicals; Hypertension -- Periodicals; 616.132005
• DOI: 10.1097/01.hjh.0000468563.12018.96
• Languages: English
• ISSNs: 1473-5598

Abstract

Abstract : Objective: Increasingly often in biomedical sciences large diverse datasets are being gathered. These datasets often include some standard clinical measurements dictated to an extent by the disease or mechanism of interest, targeted measurements of putative agents of said disease/mechanism, and widespread and untargeted so-called 'omics' data. We present two alternative analysis pipelines using a subset (n = 282) of the InGenious HyperCare cohort, focusing on left ventricular mass index (LVMI) which is an intermediate cardiovascular phenotype associated with the development of heart failure. Design and method: 17 clinical variables and 1605 molecular variables of various classes of biomolecules (microRNAs, peptides, proteins, metabolites) from blood and urine samples were analysed using simple and multiple linear regression. A novel method was employed to infer levels of 109 proteins from the 1340 urinary peptides detected by GC-MS which were also included. Our first approach involves a screening phase to detect candidate predictors, and a modelling phase using all statistically significant results as putative predictors. The second approach relies on principal components analysis to reduce the dimensionality of the dataset. Results: The first approach shows that a similar amount of data is explained by our clinical model as in the molecular model (adjusted R 2 = 0.209; 0.203), however a model using both sets of data is more effective (adjusted R 2 = 0.333). The final 'combined' model consists of four clinical variables (systolic blood pressure, heart rate, sex, history of congestive heart failure) and seven molecular variables (CO1A2, EFNA1, HCN4, PTGDS, phenylacetylglycine, an unidentifiable metabolite and a generalised measure of lipids), where for each p < 0.05. Twelve varimax-rotated principal components were found to be significant and each resembles the results of the first approach. Conclusions: The two methods both identify known biomarkers and putative biomarkers for future validation. The first method is time-intensive and results in variables in a complex network of correlations (Figure), however it results in detailed statistics comprehensively describing a large number of molecules. The second method is less time-intensive and is appealing in its inherent description of relationships between the variables, however it appears to be more prone to false positives. Figure. No caption available.