P1-17 The treelet transform—a novel method for determining patterns in adipose tissue fatty acids

• Published in: Journal of epidemiology and community health (1979) Vol. 65; no. Suppl 1; pp. A71 - A72
• Main Authors: Dahm, C C, Østergaard, J N, Gorst-Rasmussen, A, Jakobsen, M U, Schmidt, E B, Tjønneland, A, Overvad, K
• Format: Journal Article
• Language: English
• Published: London BMJ Publishing Group Ltd 01.08.2011; BMJ Publishing Group LTD
• Subjects: Adipose tissue; adipose tissues; Cancer; Diets; Fatty acids; Health risks; obesity; Principal components analysis
• ISSN: 0143-005X; 1470-2738
• DOI: 10.1136/jech.2011.142976c.11

IntroductionDietary fatty acid intake may be associated with risk of obesity and non-communicable disease. Adipose tissue fatty acids are correlated, reflecting shared dietary sources and metabolic processes. To date fatty acids have been investigated individually, or using principal component analysis (PCA), but interpretation of such studies is not trivial. The treelet transform (TT) is a novel method for generating sparse factors that describe the correlation structure of the data. In studies of dietary patterns TT is as efficient in extracting factors as PCA, and simpler to interpret. We therefore compared factors determined by PCA and by TT to evaluate interpretability of patterns in adipose tissue fatty acids.Methods34 fatty acids from adipose tissue biopsies were determined in a random sample of 1100 men and women from the Diet, Cancer and Health study. PCA and TT were conducted on the fatty acid data correlation matrix. The stability of the analyses was evaluated, and the highest variance factors were extracted and descriptively compared.ResultsTT factors consisted of distinct groupings of 3–8 fatty acids, generally characterised by hydrocarbon chain length and saturation status. PCA factors consisted of complex weightings of all 34 fatty acids, where some fatty acid groupings loaded strongly on some factors.ConclusionsFatty acid patterns determined using TT are considerably simpler to interpret than those generated by PCA, an advantage in studies of the effects of complex multi-dimensional exposures. Future work will relate these patterns to risk of disease.