Skip to main content
Figure 3 | BMC Bioinformatics

Figure 3

From: Query-based biclustering of gene expression data using Probabilistic Relational Models

Figure 3

Effect of noisy seed genes on bicluster results. Analysis of biclusters, obtained by the different algorithms after biclustering in the presence of an increasing number of random genes (1 = 20%, 2 = 40%, 3 = 60% and 4 = 80% of random genes) added to the true seed set of one seed set of high quality (panel a: ArcR_FadR) and one seed set of low quality (panel b: NagC). The procedure of adding random sets was repeated 100 times for each seed set. Obtained biclusters were categorized according to their behavior with respect to the true seed genes, using the categories ‘Empty’ biclusters, ‘Only full seed set’ biclusters (containing all initial seed genes but without additional genes) and ‘Only part of the seed set’ biclusters (containing only a part of the initial seed genes but without additional genes), ‘Only (part of) random seed set’ biclusters (containing only (a part of) the random genes that were initially added to the true seed set), ‘(Part of) random seed set and additional genes’ biclusters (containing a (part of) the random genes that were initially added to the true seed set and extra genes that were not a part of the initial seed set), ‘Full seed set and additional genes’ biclusters (containing all seed genes together with additional genes), ‘Part of the seed set and additional genes’ biclusters (containing a part of the initial set of seed genes together with additional genes), and ‘Drift away’ biclusters (clusters containing one or more genes, but none of the seed genes).

Back to article page