One-step and two-step bait fishing: Evaluation of results

The evaluation of the SILAC AP-MS experiments required a threshold for the SILAC ratio to define when an identified protein should be considered a contaminant or putative interaction partner. Other studies apply a statistical measure such as the Grubbs outlier test [1] or the z -test [2] to measure whether the deviation of a SILAC ratio is significant. These tests presuppose that the parameter to be tested follows a standard normal distribution in the test population (or at least in the fraction considered to be background). This prerequisite is not met for most datasets produced in this study (checked by Shapiro-Wilk normality test, not shown), making these tests inapplicable. Hence, a fixed threshold for rating a prey protein as an interactor or contaminant was defined based on our observations. Manual inspection of the results revealed that in most experiments there was a clear separation of the vast majority of proteins with association scores of less than 3 from a few clearly enriched proteins with much higher association scores. For putative membrane-associated baits (e. g., proteins from the core signaling complex) and sticky baits, no clear separation between the majority of proteins and putative interactors was achieved at this asociation score level. Here the number of proteins first started to decline rapidly at an association score of around 5 to 7, indicating that higher association scores were required for membrane associated interactions. Based on these observations, the threshold for accepting an interaction was set at an association score of 7. Without a doubt this fixed threshold neither represents the correct score for all cases nor can it measure the “quality of the guess” as in a probability-based score. However, the use of a rather conservative threshold of 7 should result in a relatively low false positive rate, albeit at the expense of an elevated false negative rate. Furthermore, almost all interactions discussed in this manuscript were identified with an association score of 10 or higher and thus are highly unlikely to be false positives.