The gene-gene functional relationship can be specifically detected by kappa statistics. (a) Kappa scores were calculated for all possible combinations of human gene-gene pairs (approximately 300 million). Only gene-gene pairs with a higher number of annotation terms in common possibly have good kappa values. The box plot consists of the smallest and largest observations at the two end points (95% confidence interval), as well as a box from the 1st to 3rd quartiles. The blue and red lines represent median and mean observations, respectively. (b) Kappa scores were calculated for all possible human gene-gene pairs, gene-gene pairs with randomized annotation terms, all collected protein-protein interacting pairs, and all 'chemokine' gene pairs, respectively. The distributions of those kappa scores from protein-protein interacting pairs (pink) and 'chemokine' gene pairs (light blue) significantly shift to the high value end compared to human total (blue); conversely, the kappa score distribution (yellow) of gene pairs with randomized annotation terms remains in the lower value end below 0.35. Interestingly, for the human genome (blue), over 50% of the kappa scores equal 0 (no detectable relationships) and >95% are lower than 0.35. Altogether, this indicates that kappa statistics can specifically detect the gene-gene functional relationships.
Huang et al. Genome Biology 2007 8:R183 doi:10.1186/gb-2007-8-9-r183