Figure 5
From: Mining gene expression data by interpreting principal components
Diabetes PC14 Sample Partitioning is Correlated with Certain Covariates. When sufficient covariate data is available (a number of the measurements are missing for certain covariates), covariate distributions are compared across partitions and significant differences are reported (as in Table 7). When a covariate is identified as significantly correlated with a principal component's sample partitioning, covariate distribution plots can be generated to further investigate and evaluate the apparent relationship. For example, the diabetes dataset's PC14 extreme genes partition the samples into UP 14 (n = 8), FLAT 14 (n = 17) and DOWN 14 (n = 10) based on their expression patterns. PC14's UP 14 vs. {FLAT 14 +DOWN 14 } sample partitioning appears related to the Insulin_0 measure (sig = 0.001), and the {UP 14 +FLAT 14 } vs. DOWN 14 partitioning appears related to Type2b_(%) (sig = 0.008). The mean expression of the PC14EG-high genes appears modestly correlated with Insulin_0 (r = 0.411) and with Type2b_(%) (r = 0.467). UP 14 samples are in red, FLAT 14 are black, and DOWN 14 are blue.