Skip to main content

Table 3 Comparing the biological relevance and novelty of different biclustering solutions

From: BicPAM: Pattern-based biclustering for biomedical data analysis

Dataset Approach Bics Avg.Genes ×Conds Bics sig.enriched Coverage and exclusivity of enriched GO terms
dlblc BicPAM 56 83 ×7 43 (77%) Highest number of exclusively enriched terms (partial list in Table 4).
(human BiModule 322 62 ×4 79 (25%) Absence of closing options leads to redundant and less significant terms.
genome) DeBi 31 73 ×6 21 (68%) Loss of relevant terms due to the inability to discover all maximal biclusters.
  CC 10 41 ×33 5 (50%) Exclusive bicluster related with circulatory & cardiovascular system development.
  ISA 72 23 ×8 8 (11%) Exclusive bicluster for extracellular structure organization and heparin binding.
  Plaid 3 12 ×49 1 (33%) Majority of genes modeled in a single background bicluster with general terms.
  Fabia 10 79 ×35 6 (60%) Small bicluster with superior enrichment of antigen binding functions.
  Bexpa 10 16 ×87 2 (20%) Small sets of genes supported by large number of conditions.
  Samba 100 17 ×6 18 (18%) Dedicated terms for antigen processing, peptide cross-linking and disassembly.
  OPSM 12 128 ×5 5 (42%) High variance of genes and conditions; some of the biclusters with low genes (coherency across high conditions) have exclusive significantly enriched terms.
hughes BicPAM 47 360 ×7 38 (81%) Exclusive enriched terms due to flexible coherency and post-processing criteria.
(yeast BiModule 219 285 ×4 43 (20%) Terms with lower sig. than terms from noise-tolerant BicPAM solutions.
genome) DeBi 28 317 ×7 21 (75%) Terms observed across very small sets of conditions (≤5) are not enriched.
  CC 10 228 ×58 6 (60%) GO terms covered by BicPAM constant biclusters.
  ISA 8 120 ×4 5 (63%) Small biclusters with exclusive significance GO terms: spindle pole and karyogamy.
  Plaid 8 78 ×39 3 (38%) One bicluster with higher significance for fungal-type cell wall assembly.
  Fabia 10 210 ×49 5 (50%) Higher significance observed for actin cortical patch and oxidoreductase GO-terms.
  Bexpa 72 42 ×49 1 (10%) Low number of enriched terms (probably due to the low genes per bicluster).
  Samba 120 18 ×9 11 (9%) Enriched terms covered by pattern-based biclustering solutions.
  OPSM 6 531 ×4 3 (50%) Exclusive bicluster for the negative regulation of metabolic processes.
gasch BicPAM 149 411 ×8 123 (83%) Large diversity of highly significant GO-terms (partial list in Table 4).
(yeast BiModule 653 287 ×4 159 (24%) Large but incomplete set of GO-terms as it excludes non-constant biclusters.
genome) DeBi 82 310 ×6 61 (74%) Significance of terms slightly differ than BicPAM due to the handling of noise.
  CC 10 203 ×79 7 (70%) Enriched terms appear in BicPAM solutions with higher significance.
  ISA 23 292 ×22 18 (78%) Enriched terms covered by pattern-based biclustering solutions.
  Plaid 6 48 ×12 3 (50%) Biclusters (apart from background layer) with lower enrichments than peers.
  Fabia 10 310 ×41 8 (80%) Bicluster with higher sig. for specific proteasome complexes.
  Bexpa 10 63 ×29 3 (33%) The few biclusters with deviation in size (higher genes) are significant.
  OPSM 16 212 ×8 11 (69%) One bicluster with higher significance for pre-ribosome functions.