To obtain relative affinity estimates that are both accurate and precise, we integrate information from multiple rounds of selection using LOESS regression. (A) Direct comparison between the normalized fold-enrichment from R0 to R1 and the normalized square root of the fold-enrichment from R0 to R2 of all 12-mers for Exd-Lab using the data from Slattery et al. (2011). The deviation from the straight line is presumably due to a combination of binding saturation and build-up of PCR bias. These effects are expected to be less severe in the earlier round, and therefore R1/R0 is a more accurate predictor of relative affinity. However, since counts are lower in R1 than in R2, the value of R1/R0 is also less precise. The error bars denote the standard error in the estimate of the relative affinity, calculated as described. (B) Comparison between the R1-based affinity estimates and LOESS-based estimates resulting from integration of R1 and R2 data.