Skip to main content

Table 3 Comparing performance of our method (MCIP) with and without the use of priors to k -means clustering

From: Identifying pathogenic processes by integrating microarray data with prior knowledge

  Heart failure data Melanoma data
  Sens. Spec. PPV AUC K Sens. Spec. PPV AUC K
MCIP with priors 0.62 0.68 0.023 0.65 2 0.67 0.6 0.030 0.64 18
MCIP without priors 0.61 0.64 0.020 0.62 3 0.68 0.53 0.024 0.60 20
Kmeans 0.52 0.65 0.019 0.59 3 0.70 0.48 0.021 0.58 10
  1. The performances are obtained by evaluating the result clusters against a literature based reference network comprising pairs of genes co-cited in Pubmed articles with the Medical Subject Headings (MeSH) left ventricular hypertrophy (for the heart failure data) and melanoma cancer (for the melanoma cancer data). Sens.=sensitivity, Spec. = specificity, PPV = Positive predictive value, AUC = Area under receiver operator curve, and K = number of clusters, which is automatically found by our method, and by using the Gap index for k-means clustering.