In the NanoTemper Technologies software products that analyze MST data (applicable to the following software versions and earlier: MO.Control 1.6.1, MO.Affinity Analysis 3.0.5, MO.Screening Analysis 1.0.3), different quality values are returned to the user to judge how well a selected fit model matches the measured raw data. Some values can also be used to assess the quality of the measured data altogether.

Response Amplitude

𝑅𝑒𝑠𝑝𝑜𝑛𝑠𝑒 𝐴𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒= |𝑢𝑛𝑏𝑜𝑢𝑛𝑑−𝑏𝑜𝑢𝑛𝑑|

Where unbound and bound are the respective estimated values from the fit. “Unbound” is the plateau at very low concentrations of ligand (also called baseline), while “bound” is the plateau at very high concentrations of ligand (also called saturation).

Kd Confidence

In general confidence intervals describe the certainty of a fit. We calculate the confidence interval of the Kd from the variance of the fitted parameter. This variance is directly derived from our fit-algorithm (Levenberg-Marquardt). With a confidence of 68%, Kd is within the given range. The lower this number, the more confident one can be about the given Kd.

Standard Error

Is also called RMSE (root mean squared error) in literature. In MO.Affinity Analysis it is defined as

With the residual degree of freedom 𝜈=𝑛−𝑚; n is the number of data points and m is the number of parameters that are fitted (four parameters for both, Kd- and Hill-model, except any parameters are fixed). The y-values of the fitted curves at position i are denoted by 𝑚𝑖 and the actual data points are denoted by 𝑦𝑖.

Reduced χ2

This value is only calculated for merge sets that contain two or more replicates.

Where 𝑚𝑖 denotes the y-values of the fitted curve, 𝑦𝑖 denotes the averaged raw-data y-values and 𝜎𝑖 denotes the standard deviation of the averaged raw-data y-values.

The reduced χ2 is then defined as

With the residual degree of freedom 𝜈=𝑛−𝑚; n is the number of data points and m is the number of parameters that are fitted (four parameters for both, Kd- and Hill-model, except any parameters are fixed).

In MO.Affinity Analysis the reduced χ2 can become quite large. The reason for this is that replicates are often very similar. This yields a small standard deviation. Since we divide by these small values, the number can become quite high. Therefore, the absolute value of the reduced χ2 alone is not a useful parameter from which to judge data quality. It is however very useful for comparing data quality between replicates or comparable samples. In such cases, the smaller χ2 for one particular dataset in comparison to other datasets, the better the data quality.

Signal to Noise

The signal-to-noise is calculated by dividing the response amplitude by the noise. The noise is calculated as the standard deviation of the residuals from the fit.

Where ri denotes the residual of the fit at a given data point and 𝑟̅ the average of all residuals. The number of data points is given by n.

The signal to noise is a good parameter to judge data quality. A value of more than 5 is desirable while a value of more than 12 corresponds to an excellent assay.