Figure 1

Cluster statistics A: Histogram showing the log number of clusters as a function of log cluster size, based on the clustering at Pearson correlation coefficient 0.75 cut-off. Numbers on the x axis denote cluster size intervals (2), (3–4), (5–8), (9–16),... B: Co-expression as a predictor for shared function, protein interaction and paralogy. We identified all gene pairs that correlated above or below a threshold T (X-axis). We measured the fraction of such pairs for which there was (i) a BIND database protein-protein interaction recorded in human, (ii) at least one shared gene ontology term, and (iii) evidence of paralogy. We then computed the relative probability for genes above T with this feature, compared to gene pairs below T. At expression correlation 0.80, co-expression was associated with a 100-fold relative probability for genes to encode protein interactors, a 10-fold probability for genes to share functional annotation, but only a 3-fold probability for genes to be paralogs. C: Fraction of clusters with at least one over-represented GO term (Y axis), as a function of cluster size (X axis). GO term over-representations were computed at a 10% false discovery rate.