Reply to "Exhaustive benchmarking of the yeast two-hybrid system"

Braun et al. reply:

We applaud the thorough and revealing study by Chen et al.¹ in this issue of Nature Methods. The work expands our previous findings in thoroughly characterizing different yeast two-hybrid (Y2H) implementations, with respect to overall assay sensitivity, by testing each implementation against a panel of reference protein-protein interactions². The standardized reference sets³ make the data easily comparable to our previous analyses and clearly demonstrate that different Y2H assays detect different subsets of interacting pairs of proteins^2,3. Given the proven utility of using several assay configurations, the next question is where and how to deploy them. The high-throughput capabilities of Y2H^4,5 make it an ideal primary screening assay. Having multiple versions of Y2H that detect different subsets of interactions will be of a great value to generate more comprehensive data sets, which would then need to be validated using a scheme such as the “confidence scoring” scheme that we proposed². A key concept of our confidence scoring method is that any interaction detected by a given screening assay is subsequently confirmed by multiple orthogonal validation assays. The screening and validation assays must be as independent from each other as possible to eliminate the danger of systematic assay-dependent artifacts, which could make protein pairs appear as robustly interacting when they may not be. Use of a single type of assay for both screening and validation, even if implemented in different configurations, may introduce such systematic biases. It is therefore critical to obtain orthogonal validation ideally of all interacting pairs identified in an initial screen.