Systematic interpretation of genetic interactions using protein networks

Abstract

Genetic interaction analysis,in which two mutations have a combined effect not exhibited by either mutation alone, is a powerful and widespread tool for establishing functional linkages between genes. In the yeast Saccharomyces cerevisiae, ongoing screens have generated >4,800 such genetic interaction data. We demonstrate that by combining these data with information on protein-protein, prote in-DNA or metabolic networks, it is possible to uncover physical mechanisms behind many of the observed genetic effects. Using a probabilistic model, we found that 1,922 genetic interactions are significantly associated with either between- or within-pathway explanations encoded in the physical networks, covering approximately 40% of known genetic interactions. These models predict new functions for 343 proteins and suggest that between-pathway explanations are better than within-pathway explanations at interpreting genetic interactions identified in systematic screens. This study provides a road map for how genetic and physical interactions can be integrated to reveal pathway organization and function.

Figure 1

Method overview. A combined physical and genetic network is searched to identify between- or within-pathway models of genetic interactions. The between-pathway model implies two groups of proteins (pathways) with many physical connections within each pathway (solid blue links) and genetic interactions spanning between pathways (dotted red links). The within-pathway model implies many physical and genetic interactions within the same group of proteins. In the search, 360 and 91 network models were identified that correspond to between- or within-pathway searches, respectively.

Figure 2

Between-pathway explanations for genetic interactions. (a) Several high-scoring models are shown (M,N,T; Q,V; O,U,Y). Blue solid and red dotted links indicate physical and genetic interactions, respectively. (b) Bird’s-eye view of all between-pathway models obtained from a search on a reduced network composed of SGA and DIP interactions. Each node [A]–[Z] represents a physical pathway; groups of genetic interactions between pathways are condensed into a single link called a ‘bundle.’ Node colors indicate significant Gene Ontology annotations. Solid gray lines connect pathways that share one or more proteins; such pathways may represent different components of a larger mechanism.

Figure 3

Within-pathway explanations for genetic interactions. A total of 91 pathways were identified, of which four examples are displayed. Color is used to indicate the data set from which each interaction was drawn.

Figure 4

Genetic interaction prediction schemes. Two different schemes are proposed for predicting genetic interactions, depending on the underlying network model. Observed genetic interactions are shown in red, while the corresponding predicted genetic interactions are shown in gray. (a) Under the between-pathway model, two incomplete bipartite motifs are shown which predict a genetic interaction between genes b and b′. (b) Under the within-pathway model, common genetic neighbors are used to predict a genetic interaction between genes d and d′. Note that these diagrams contain additional incomplete motifs which have been omitted for clarity: the motifs in a can be rearranged to predict genetic interactions (a to c′) and (c to a′); the motifs in b can be rearranged to predict (e to f).

Figure 5

Success rate of genetic interaction prediction versus the stringency of prediction. Success rate is measured through cross-validation as (predicted positives)/(predicted positives and negatives). Stringency is defined by the minimum number of incomplete bipartite motifs required for prediction. Blue diamonds mark the success rate for predictions in which incomplete motifs must occur in a between-pathway model. The success rate is dramatically higher than for naive predictions (magenta) which predict interactions in the same manner, but are not constrained by the physical network. Even for much more stringent prediction criteria, the success rate of naive predictions fails to exceed that of the between-pathway predictions (inset).