Comparative enrichment analysis of biomarker-based patient clusters: A quantitative analysis.

The number of validated terms found to be enriched in clusters generated with the different pre-processing procedures and clustering algorithms tested in this study. Validated enrichments and validated clusters are defined by the recurrence of statistically significant enrichment in the training and validation datasets. The clusters that were generated from the test dataset using a particular pre-processing clustering combination were subjected to enrichment analysis with 19 health/lifestyle labels (i.e. searching for statistically significant over-representation of patients with the trait in each cluster). An artificial neural network, trained with the cluster assignment of each individual in the training dataset, was used to classify individuals from the validation dataset using the same clinical biomarkers subjected to the same pre-processing algorithm as was the test dataset. The resulting clustering of the validation set was also subjected to enrichment analysis with the same terms as was the training set. An enrichment was deemed to be a validated enrichment if the same label was enriched in the test and validation datasets. A validated cluster was defined as a cluster sharing at least one enriched term between the test and validation sets (i.e. the number of clusters enriched in the training set). The enrichment factor for each pipeline is the average enrichment factor of the three most significant enrichments. K-mns = K-means; NoramTranf = transformation to normal; Z-score = Z-score normalization; AgeAdj = age adjustment followed by Z-score normalization.