Methods and Techniques Discussion => Enzyme Linked ImmunoSorbentAssay (ELISA) => Topic started by: ahiscox on November 29, 2007, 10:40:33 AM

I've been doing ELISAs for several years as a grad student, and I've always used a linear best fit line for my standard curve. I'm currently trying to troubleshoot an ELISA that someone else has done, and noticed that he was using a 4parameter standard curve. When I do a linear line on all of his data, I get rsquared values that are a bit lower than his 4parameter best fit lines.
Is a 4parameter best fit line OK to use? I've always been told to use a linear line, but when I use a linear line on his raw data, I get slightly smaller rsquared values (.9978 vs. .9789 for example).
Thanks.

HI there,
You can certainly use a 4parameter best fit. But I think it depends on what the company you order the product from suggests. I order my ELISA kits from R & D Systems (Duoset) and was recommended to use a 4parameter fit. I think is important to follow the company's instructions to get the best results! Good Luck!

For future reference, check out this stepbystep guide for generating a standard curve in Excel.
http://www.mdbioproducts.com/resources/protocols/standardcurve (http://www.mdbioproducts.com/resources/protocols/standardcurve)

For ELISA standard curves, the 4 parameter logistic (4PL) (http://www.miraibio.com/blog/2010/08/the4parameterlogistic4plnonlinearregressionmodel/) is definitely better than a linear fit. A linear fit has no limits (i.e. maximum and minimum asymptotes) which does not accurately describe biological systems. I would even go a step further and say that the 5 parameter logistic (5PL) (http://www.miraibio.com/blog/2009/02/5pllogisticregression/) is an even better fit than the 4PL because the 4PL assumes symmetry around the inflection point where the 5PL does not. In addition, I would not recommend using R^2 to compare linear vs. nonlinear fits because it is like comparing apples to oranges. The R^2 is good for comparing linear fits to each other but it is not accurate for nonlinear regression curves like the 4PL and 5PL. Instead, I would recommend using the root mean square error (RMSE) (http://en.wikipedia.org/wiki/Root_mean_square_deviation) as a goodness of fit.
Here are some blog posts for those that are curious in learning more about the 4 parameter logistic (4PL) curve fit (http://www.miraibio.com/blog/2010/08/the4parameterlogistic4plnonlinearregressionmodel/) and/or the 5 parameter logistic (5PL) curve fit (http://www.miraibio.com/blog/2009/02/5pllogisticregression/).
Since we are on the topic, here is a blog post for Tips for ELISA Data Analysis (http://www.miraibio.com/blog/2009/06/tipsfordataanalysis/).
I hope this information helps.
Allen Liu
* Disclaimer: I am a Hitachi Solutions employee.