Plot describing metabolite importance and robustness in predicting visceral fat adiposity as assessed by Random forest analysis using metabolic data collected at V0 and V2.

<p>Visceral adiposity was associated with increasing concentrations of amino acids (glutamine, leucine/isoleucine, phenylalanine and tyrosine), lysophosphatidylcholine LPC 24∶0 and diacyl phospholipids (PC 30∶0, PC 34∶4). In addition, visceral adiposity was marked by a depletion in ether lipid species PC<i>-O</i> 36∶3, PC<i>-O</i> 40∶3, PC<i>-O</i> 40∶4, PC<i>-O</i> 40∶6, PC<i>-O</i> 42∶2, PC<i>-O</i> 42∶3, PC<i>-O</i> 42∶4, PC<i>-O</i> 44∶3, PC<i>-O</i> 44∶4, PC<i>-O</i> 44∶6, and two diacyl phosphocholines (PC 42∶0 and PC 42∶2). To reflect the weight of the selected biomarkers in the classification of visceral adiposity, a pooled mean decrease of accuracy for each compound was calculated from 10000 forest generations. Higher variable importance corresponds to higher values of pooled mean decrease in accuracy. Key: IPVF, intraperitoneal fat volume; LPC, Lysophosphatidylcholines; PC, Phosphatidylcholines; PC-O, 1-O-alkyl-2- acylglycerophosphocholines; Ratio1, intraperitoneal/subcutaneous fat ratio; Ratio 2, intraperitoneal/abdominal fat ratio. Assignment of PC-O species is made on the assumption that only even numbered carbon chains are present. A potential overlap between PC species containing odd-chain fatty acids and even-chained PC-O species cannot be excluded with low mass resolution.