Abstract #4

Evaluating quantitative measures of epistasis for predicting functional relationships. Ramamurthy Mani¹, Robert P. St Onge², Julia Oh², Michael Proctor², Eula Fung², Ronald W. Davis², Corey Nislow², Guri N. Giaever¹, Frederick P. Roth¹. 1) Harvard Medical School, Boston, MA, 02115; 2) The Department of Biochemistry, Stanford University, Stanford, California, USA, 94305.
   Assignment of gene function is the central problem of functional genomics. The strategy of predicting function by first predicting functional linkage has been proven useful. Genetic interaction or epistasis between two genes allows prediction of functional linkage: Synthetic sick or lethal interactions are particularly valuable for this purpose (Tong et al., 2004), being more sensitive predictors of functional linkage than protein interactions (Wong et al., 2005). We examined a systematic study of genes involved in DNA repair, in which fitness was quantitatively measured for all possible single- and double-deletion strains involving 26 gene deletions (St Onge et al.). We evaluated several alternative quantitative measures of epistasis, e.g., deviation from multiplicative expectation (epsilon) and epsilon profile correlation, for their ability to predict functional linkage. Logistic regression analysis showed that a combination of epistasis measures gave the best performance. One sub-class of alleviating interactions, which we term ‘co-equality’, corresponded closely with protein complexes that function as cohesive units. Within this DNA repair-focused data set, alleviating interactions were more predictive of functional linkage than aggravating (or synthetic) interactions, suggesting an increased focus on such interactions in future large-scale studies.

Return to YGM 2006 Home at SGD