Evaluation of logistic and polynomial models for fitting sandwich-ELISA calibration curves

Appropriately modeled calibration curves are important for accurately estimating the concentrations of proteins in samples evaluated in sandwich-format enzyme-linked immunosorbent assay (ELISA). Calibration curves are commonly fit using polynomial or logistic models. We compared the fit of a quadratic, cubic and 4-parameter logistic model for highly-replicated calibration curves across seven assays used for quantifying transgenic proteins in commercial crops. Results indicate that it is typically undesirable to include zero-concentration data when modeling these curves over the quantitative range, and simple polynomial models are typically preferable to the commonly recommended 4-parameter logistic model. These results are applicable to assays where precision constraints preclude interpolating results from the flat portions of the calibration curve, and it is under these conditions that the moderate improvements in accuracy described here will have impact.