Replicates and RSD% on fitted parameters

Hi everyone!
I am measuring an Antigen-Antibody interaction, and i am doing replicates. The problem is that the RSD% is too high! I have done this experiment with many samples, and it always happens the same. Do you know if it is an accepted error for the replicates?

Here it is an example. The results after a 1:1 fitting were the following:

Experiment 1:
kon: 4.59e5 M-1s-1; SE: 0.02e5 M-1s-1
koff: 6e-5 s-1; SE: 2e-5 s-1
KD=koff/kon: 1.3e-10 M

Experiment 2:
kon: 4.63e5 M-1s-1; SE: 0.06e5 M-1s-1
koff: 14e-5 s-1; SE: 2e-5 s-1
KD=kon/koff: 3.1e-10 M

KD
Average: 2.2e-10 M
SD: 1.2e-10 M
RSD%=(SD/Average)*100: 55.7%!!!!!!

Thank you very much!
Sincerelly
Pablo

Hi Pablo,

May be this publication is of help.

Rich, R. L., Papalia, G. A., Flynn, P. J., et al.; A global benchmark study using affinity-based biosensors. Analytical Biochemistry (386) 194-216; 2008.

In Table 1 the rate and dissociation constants of the study are presented.
This are constants averages over several participants and the RSD% is sometimes low, sometime high.

However, I think you should not give to much weight to the value. I agree smaller is better, but with two experimental values you can be unlucky.

Kind regards

Arnoud

Thank you very much Arnoud!
I checked the paper you sent me. I see that exist quite a variability over different surfaces and equipments.

Do you think it is correct to inform the average+/-SD? Avoiding the RSD%. I would like to add a third repetition of the assays, but it is time and sample consuming! I followed the recomendations given in SPR Pages that two repetitions were acceptable.
Again, thanks a lot!!!
Sincerelly
Pablo

Hi Pablo,

I would report a value and SD because this is the standard way to do it. The RSD% (or CV) is more to see if the replicates are close to gether.
As for replicates, I think two is the minimum. For robust kinetic values you should test different batches of ligand and analyte etc.....

Kind regards

Arnoud

Thank you very much for yor reply, Arnaud.

I have been reading about constants statistic distribution. It appears that KD follows a log-normal ditribution ("Assessing the distribution of parameters in models of ligand-receptor interaction: to log or not to log". Christopoulos A. TiPS Vol 19 351-357. September 1998. And "The Universal Statistical Distribution of the Affinity, Equilibrium constants, Kinetics and Specificity in Biomolecular Recognition". Zheng X, Wang J. PLOS Computational Biology 2015) So the correct way to inform the data is as a geometric mean of the values, together with a geometric SD.

I am interested in the scatter of the data. I tried calculating the pKD=-logKD, and then calculate the CV of the pKDs replicates. By doing this, i get CV%<5%. Do you think this may be a correct way to analyze the data scatter, considering the log-normal distribution of KD?

Thanks a lot!
Best regards

Pablo

Hi Pablo,

I am not a statistician, but it think you should just report a value + SD.
I my opinion you are seeking the smallest error you can report, but your values are what they are. I you want a more defined value you should repeat the measurement but there are no guarantees that you will get a smaller SD.

Kind regards
Arnoud