Performance comparison, local methods. Methods are listed, from bottom to top, roughly in order of improving performance in predicting synthetic lethal genetic interactions. Shared metabolites indicates raw counts of shared metabolites, and "no hubs" indicates that the most highly connected metabolites were removed before calculating sharing. The Poisson score, hypergeometric p-value, and Bayes score are based on the number of shared metabolites, but do not use the full metabolite degree distribution. The method "Poisson score (met. deg.)" uses the metabolite degree distribution for an additional correction. Performance is assessed using synthetic lethal gene pairs as known positives, and non-synthetic lethal pairs as known negatives. (A) Receiver operator characteristic (ROC) curve. (B) Precision-recall (PR) curve.