Figure 3

A heatmap of sample-lipid class combinations. Given the matrix partition factor k = 4, we ran the NMF algorithm 100 times to generate consensus clustering of the combinations of samples and lipid classes. Each run resulted in a 40 × 40 connectivity matrix with an entry of 1 if sample-lipid class combinations i and j cluster together and 0 otherwise, where i, j = 1,..., 40. The consensus matrix is the averaged connectivity matrix obtained over the 100 runs. Each block along the main diagonal represents a consensus cluster of the combination of samples and lipid classes. Meta-fatty acid cluster characteristics are to the left, specific meta-fatty acid groups to the right: F = female; M = male; GL+ = with a germline; GL- = with no germline; CL = cardiolipin; CE = cholesterol ester; DAG = diacylglycerol; FA = free fatty acid; LY = lysophosphatidylcholine; PC = phosphatidylcholine; PE = phosphatidylethanolamine; PS = phosphatidylserine; SP = sphingomyelin; TAG = triacylglycerol. Only Lysophosphatidylcholine from males and females map to different locations in the matrix (*).