Choose transformations and weighting

Choose the scale used for fitting

On Model, inspect Dose transformation or Concentration transformation and Response transformation. These settings change the calculation, not just the graph. Responses are transformed before fitting, residuals, weighting and uncertainty are calculated.

Untransformed retains the numerical value. Base-2 logarithm, Natural logarithm and Base-10 logarithm retain their distinct base and require strictly positive finite inputs. Provenarium does not add an offset, take an absolute value or replace zero to make a logarithm possible.

Interpret the transformed model

Parallel-line potency is linear on log dose; a straight line on raw dose would not give the same response-independent potency ratio. The untransformed-dose sigmoid potency option retains a Hill curve in positive original dose units. It does not permit zero or negative doses.

Log-base changes alter the units of fitted slopes and locations. Read the saved settings before comparing parameters. Relative potency and concentration outputs are returned to their declared original units; parameters on the fitted scale and residuals still describe the chosen transformed axes.

4PL symmetry describes its logistic coordinate and fitted response scale. A nonlinear back-transformation can change how symmetric the curve looks on the original axes. A log concentration transform alone says nothing about whether response variance is constant or changes with signal.

Match Weighting to a justified variance assumption

Here m is the model’s expected response after the selected response transformation. It is updated during fitting; these are not weights computed from each observed response. Choose from prior assay knowledge or a justified variance assessment, not by which choice produces a preferred potency or passing criterion.

Inverse expected response requires positive fitted predictions. Inverse squared expected response requires nonzero fitted predictions and finite weights. Relative potency also requires positive transformed observations for the first option, and nonzero observations of one sign with predictions of that sign for the second. Thus a log-response potency analysis with inverse expected response needs raw responses above 1; inverse squared weighting needs raw responses consistently above or consistently below 1.

Changing units can change the transformed response and its weight domain. Do not change labels, add offsets or switch weights merely to suppress a domain failure. Investigate the scientifically intended values and model.

Recognize an invalid fit

If values are incompatible with the selected transformations or weights, or the curve cannot be fitted reliably, Provenarium reports a calculation failure. It keeps your selected settings and observations. Check for input errors, then follow your procedure for investigating or repeating the assay. Keep the failed result with your records.

• Scientific background: EDQM CombiStats 7 manual, sections 7.2–7.6