Defining the functional relationships between proteins is one of the important tasks in the post-genomic era. A classical approach to understand gene functional relationships is producing combination mutant in two genes to observe genetic interactions [1
]. Genetic interaction refers to the phenomenon in which the combined effect of mutations of two genes differs from individual effects of each mutation [2
]. In the extreme cases, mutation of two nonessential genes could lead to a lethal phenotype. This kind of genetic interactions is referred as synthetic lethal genetic interactions (SLGIs). The genome-wide SLGIs have attracted much attention as they are capable of defining the genome-wide functional relationships between proteins, pathways and complexes [2
]. The SLGIs also have potential for finding drug target or drug combinations [5
Representing the structures and functions of proteins, protein domains are usually regarded as building blocks of proteins and are conserved during evolution. The mutation of a gene causes the loss of function of its protein product, which may accredit to the loss of protein domains in the protein product. Then, the effect of the mutation of two genes may be caused by the loss of protein domain combinations in both protein products. We refer the phenomenon in which combined effect of mutations of two domains in two proteins differs from individual effects of mutation of each domain as domain genetic interactions. The domain genetic interactions may correlate to SLGIs. We may use the domain genetic interactions to explain and predict the SLGIs. Furthermore, in multidomain proteins, different domains may fulfill different functions independently or collaboratively. Although genetic interaction analysis provides a promising method to understand the functional relationship between proteins [6
], we cannot tell the contributions of different domains to certain functionality of multidomain proteins based on their genetic interactions. Studying the domain genetic interactions may help elucidate the SLGIs between multidomain proteins at domain level.
Genetic interactions are usually identified by mutant screens [7
]. Recently, high throughput technologies, such as synthetic genetic arrays (SGA) [8
] or synthetic lethal analysis by microarrays (SLAM) [9
], have been developed for parallel and massive detection of genetic interactions. However, even with high throughput methods, experimental discovery of SLGIs is still overwhelming. Therefore, it is of interest to computationally predict SLGIs. Several computational approaches have been proposed for the prediction of SLGIs [10
] and various features, such as protein interactions, gene expression, functional annotation, gene location, protein network characteristics, and genetic phenotype, had been utilized by these methods. Recently, Park et al. [15
] also used protein domains as one of the features to predict interactions, including genetic interactions.
In our previous study [16
], we performed a cross validation study on predicting SLGIs using protein domains as features. Our support vector machine (SVM) classifiers were able to achieve high performance with AUC (The area under the ROC curve) as 0.9272. These results suggested that using domain information may catch the genetic relationships between proteins. In this study, we first applied the Maximum Likelihood estimation (MLE) approach to estimate the probabilities of domain genetic interactions from yeast SLGIs. The EM algorithm developed by Dempster et al. [17
] had been used to iteratively compute the maximum likelihood. Recently, the MLE method had been used to infer domain physical interactions from protein physical interactions [18
]. We identified significant domain genetic interactions, which rarely overlap with the domain physical interactions from iPfam database [20
]. We also showed that domains in multidomain yeast proteins contribute to their genetic interactions differently. Then, we were able to predict novel yeast SLGIs using probabilities of domain genetic interactions, which demonstrate the ability of protein domains in predicting SLGIs. We were also able to apply our predicted yeast SLGIs to understand the compensatory pathways. A complete description of our results and methods is given in the sections below.