Hello community,
I'm currently working on a fluorescence based assay and experience recovery issues in my samples at low analyte levels. I noticed big residual deviations from the lowest calibration data point. That's why I want to optimize my calibration curve and force it through the lowest calibration level to minimize this problem and get more realistic values. I'm aware of the fact that this is far away from good scientific practice but it's better than calculating negative concentration values above the lowest standard point, completely distorting my results.
Of course an additional low-level calibration curve would have been the better option... but I can't repeat these past experiments as I have a lot more to do during my thesis and I know that this approach gives me enough evidence to proceed my future experiments.
My options for calibration are MS Excel, GraphPad and R (I'm not familiar with coding yet). I can think of a tedious workaround in MS Excel by transforming the lowest data point to the origin and fixing it there using MS Excels built in option. But i'd be happier using a more elegant way to perform this.
Thank you!
I got your point that you cannot repeat the experiment.
However, your statement “I want to optimize my calibration curve and force it through the lowest calibration level to minimize this problem and get more realistic values”, as you said “this is far away from good scientific practice”.
If your is only the first screening, as I guess from your statement “evidence to proceed my future experiments”, it would be better to calculate your sample only in comparison to the lowest standard point. Just because it is a screening experiment and work with “recovery issues in my samples at low analyte levels”.
Based on your preliminary results, you will have an idea of your working concentrations and please, take in consideration the uncertainty (I guess huge) at these levels.
Roberto
PS my considerations are general because I have a few experience with fluorescence
Hello Roberto,
thank you for your answer. I applied 1-point calibration and it does give me better results than my previous linear function. Normally, I'd apply 1-point calibration only within a narrow range of my expected analyte level.
In my case I'm extrapolating up to 5000%, so I would classify this approach as invalid as well.
The tricky thing with the fluorescence assays I'm using is their non-linear behaviour. This behaviour is additionally strongly influenced by matrix effects. To overcome this problem I'm spiking my standard solutions with artificial matrix. Maybe I'll find a model that fits my data better, by using the Akaike information criterion (AIC).
Even though this is not what I'd like to hear, this topic is a critical point to include in the discussion of my thesis and I'm glad i detected this issue. Based on the gained knowledge my future experiments will produce better results.
Have a nice weekend!
Raphael
Non-linear behaviour is statistically manageable. The point is that in the future the samples' range is within the calibration's range (at least close to it).
The matrix effects can be overcome with standard additions but have a look to the paper attached for more inside.
Good luck for your work
Roberto
- Badges
- Science topic