Asymmetric cell divisions sustain long-term hematopoiesis from single-sorted human fetal liver cells

Curve fitting procedures for bioanalytical assays are based on classical linear least squares (LSE) theory. A common procedure is to select among various models and weighting factors using the R2 as a goodness-of-fit criterion. It is questionable whether R2 is the most appropriate criterion for model selection. This is compounded by an often subjective removal of outliers. In this article, statistical curve fitting and diagnostic criteria are proposed. The fitting procedure is a Box-Cox-type power transformation of the data. The optimal transformation is obtained as the one that minimises the sum of squared deviations. Potential outlying standards are screened during the diagnostics stage as those whose jackknife percent deviations exceed 20%. The main advantage of this method is that it is objective and uniformly applicable across analytical techniques. Furthermore, the optimal transformation obtained in this way is unique. The results are demonstrated by comparing the power model to the R2 approach through the statistical analysis of 2094 analytical batches for 91 projects using various analytical techniques, namely GC, HPLC, LCMS and GCMS. The results indicate that the power model is robust and that QC batch acceptance using the power model is at least as good as the current method. These results hold true across all analytical techniques. It is thus strongly suggested that curve fitting and standard outlier detection for bioanalytical assays should be based on a power model and on jackknife percent deviations method with acceptable cut-off values.