Figure 1. Relation between the frequency of interactions in the literature and the fraction of interactions confirmed by high-throughput techniques.

Interactions that are described frequently in the literature tend to be confirmed more frequently. Because one would expect that interactions are more reliable if they are repeated often in the literature, this finding indicates that a comparison with high-throughput experiments is suitable for evaluating published interactions. B. Relation between the frequencies of the interaction partners in the literature and the fraction of confirmed interactions. Published interactions are obtained from text mining approaches (iHOP) and from expert-curated data (DIP and IntAct). For both datasets, the probability that an interaction is confirmed by a high-throughput experiment decreases with increasing popularity of the interaction partners. Bold lines code for the fraction of published interactions confirmed by at least one experimental technique, thin lines code for confirmation by at least two techniques. Thus, while interactions that are frequent in the literature tend to be more reliable (Fig. 1A), interactions of proteins that are frequent in the literature tend to be less reliable. C. Popularity of interaction partners vs. fraction of confirmed iHOP interactions for different experimental techniques. The negative correlation between the probability of experimental confirmation and popularity of the interaction partners is present for all different experimental techniques.