Figure 2

The Reliability of Inference for I-BmoI. Estimates of statistical power (reliability of inference) for I-Bmol, for both selected and unselected clone populations. 'Nonrandom' associations are presumed due to the alternate hypothesis while 'random' associations are presumed due to the null. Power is computed as the expected log₂-odds-ratio for either true-positive versus false-negative (mauve) or true-negative versus false-positive (teal). (A) For the selected sequences, the true-positive ratio is strongly correlated with the posterior log₂-odds-ratio shown in Figure 1. (B) Approximate inferential power of sites in the unselected population, each of which is expected a priori to be near zero. Note that both M1 and E83, identified as putative outliers in Figure 1, are shown to clearly (and unexpectedly) violate the null hypothesis. Such violations hallmark systematic errors or experimental artifacts.