Negative concentration??

[freeradical]

I carried out an ELISA to determine the concentration of TNF-alpha in 6 different samples. A standard curve was created in Excel and by fitting the trend line, the equation of the graph was determined to solve for the concentration of TNF-alpha. However, when working out the concentration of TNF-alpha most of the values were fairly close to where I estimated they should be except the very last concentration which when worked was a negative concentration. How is this possible? Also the x-axis has a logarithmic scale but the trend line is linear with the equation below. Any help greatly appreciated!!

Sample Mean Absorbance at 450nm
1 0.821
2 0.752
3 0.424
4 0.445
5 0.433
6 0.054

Equation of graph y=0.0089x + 0.1235 (where y=mean absorbance at 450nm and x=concentration TNF-alpha (ng/ml))

[JackBean]

You can see, that theoretically, if your concentration was zero (x = 0 ng/ml), than you wil get absorbance of 0.1235. That means, that you have bad baseline (that is, your blank sample is actually not blank, but absorbs), but more likely Excel just found the best fit, which just starts a little bit above zero 😉 So, you can choose to push the line to zero (so you will get something like y = a.x) and see, how much different it will be, or you will just assume, that your last sample contains no TNF-a

[JackBean]

quote freeradical:

Also the x-axis has a logarithmic scale but the trend line is linear with the equation below.

Regarding this. It’s your choice, what type of trend line you like (or better, chooce). You should see the R^2 value and also see, what is reasonable (like get something like y = a.x^4 + b.x^3 + c.x^2 + d.x + e doesn’t make much biological sense 😉 ). So, try both linear and logarithmic and you will see, what fits better

[mith]

You shouldn’t be trying to see what fits better. The model you use should reflect the physical of the system, i.e. light absorbance is linear and not quadratic or logarithmic no matter how good the line fits.

[JackBean]

Not necceserilly, e.g. we are using quadratic fit for Bradford determination of protein concentration instead of linear

[MrMistery]

@jackbean
why?

[JackBean]

Because it fits much better

[MrMistery]

but the bradford reaction is supposed to be linear. how can it fit better that way?

[mith]

fitting better isn’t a good reason for picking another model, if that were true, every regression model would be based on some fourier function that provides ~0 error.

[Author]

Posts