You’ve spent all of your lab time making the perfect SPR assay. Your data looks great, the Sensorgrams overlay like they are a single curve and then the inevitable happens; someone asks:
“How do you know it’s reliable?”
Assessing the fit
First thing to do is look at the sensorgram and the fit to it:
In the above example the data is in blue and the fit is the solid black line. So this data fit overlays the actual assay data very well so it’s expected that we’d observe confidence values that reflect this. Parameters that are commonly used to help define precision in an assay include:
Chi2 is a measure of the average deviation of the experimental data to form the fitted curve. Lower chi2 values indicate a better fit, but assessment of chi2 must be taken in the context of the binding level of the system. Basically the further away the fit is from your original data, the higher the Chi2 value will be and vice versa. As the Chi2 value is derived from the experimental data it’s hard to advise an exact number, but as a general rule of thumb ≤10% Rmax.
So using the data in the table above we have an average Rmax of 23.5 RU and therefore, we should ‘accept’ a Chi2 of less than 2.35. Out actual Chi2 is 0.00389 so we are much lower than that, so that gives us confidence in the fit.
One of my favourite values for confidence in data is T-values. These provide nice and simple feedback on the individual fit parameters. To determine these, you need the standard error (SE), which is generated for you by the BiacoreTM software. For SE you’re looking for a small a number as possible because a small SE indicates that changes in the parameter’s value would have a significant effect on the fitting (i.e., the confidence value is high).
To calculate T-values you simply divide the parameter’s value by the SE. The larger the number the better for T-values. My rule of thumb is:
- 10 = good
- 100 = great
- 1,000 = happy days
So the T-values of 251.3, 705.0 and 1,023.1 for T(ka), T(kd) and T(Rmax) give me increased confidence in the data.
Percentage coefficient of variation (%CV)
Much maligned but still a standard value that people expect to see, the coefficient of variation (CV) is defined as the ratio of the standard deviation to the mean. There are multiple guidelines for what an acceptable %CV is but a good rule of thumb is that ≤15% is good for intra-assay triplicate repeatability and ≤20% for inter-assay repeatability. Therefore, the %CV values in the table above are good for the kinetic parameters and for the T-values.
It’s worth mentioning here that %CV can be misleading, in the above table we can see that KD has a %CV of 8.44% as the numbers are close together. But as we’re working in the pM range a small change in value can have a large impact on the %CV.
For example, if I change the kd of Antibody 1.2 to 6.38E-5 from 7.38E-5, we can see that the %CV for Kd and KD is massively affected:
In real terms, the KD of Antibody 1.2 has changed from 19.63 pM to 16.97 pM but the %CV increases from 8.44% to 16.15%. I’d argue that a affinity shift of 2.66 pM is hardly biologically relevant but the mathematics behind it can play tricks on you.
As the saying goes – keep your mind open but not so open your brain drops out…
There you have it, all the information you need to defend your beautiful assay (should you ever need to) and the reason that sometimes the %CV can play tricks upon you. I really enjoy writing posts like these, so please keep asking questions and feedback on what you would like covered next.