There are three major differences between GeneMANIA and the majority of these other tools. The first difference is that GeneMANIA’s predictions are made ‘on-the-fly’. Practically, that means that the weights assigned each network change based on what networks the user selects or uploads into GeneMANIA. Another major difference is in the meaning of the link weights. In other systems, the displayed link weights can be interpreted as estimated probabilities that the linked genes ‘functionally interact’, in other words, that they share at least one annotated function. GeneMANIA link weights cannot be interpreted this way, as described in previous sections, though higher weights generally mean higher confidence. A final difference with some servers is that GeneMANIA uses a label propagation algorithm to select related genes [see (30) for the history of this methodology] rather than just considering genes that directly interact with the query list. As such, GeneMANIA considers direct and indirect paths between genes and the query list. Although direct interactions receive more weight than indirect ones, tight clustered groups of genes will often appear together as related genes by GeneMANIA owing to their larger number of indirect links.
