I run a Mg2+ dosis-response curve on phenylephrine-precontracted mouse aorta that should relax the aorta in a concentration-dependent manner (7 concentrations). I want to calculate EC50 of these curves in GraphPad prism. Currently I have done log transform concentrations - normalize response in % - non-linear regression - log agonist vs. normalized response - EC50. The problem is that I see very often the EC50 values I get are different from what I can see from the real % relaxation values. This could be due to two reasons, for the first at smaller concentrations I sometimes get the contraction, so my dose-response would look like 0, -1, -5, 10, 15, 20, 35, 40 % relaxation. Secondly, I am not sure whether to use Hill slope 1 or variable slope. Graph Pad says that with few concentrations one should use Hill slope 1, and I think 7 is quite few, so I used that, but I think I get more correct results when I use variable Hill slope. Am I doing the calculations right or there is a better way to calculate EC50 for my data?
The answer to your question depends on the type of data that you have. There are ways of estimating ED50 values with error limits for continuous data and also for discrete data (eg an all or none response). One way to start would be to learn to use the excellent statistical package called R. It will provide you with methods for analyzing your data and with methods for plotting your data. There is useful information about dose response relationships in the following reference.
https://static-curis.ku.dk/portal/files/153304510/journal.pone.0146021.pdf
It gives examples using R.
For all or none effects an author to look for from way back when is D.J. Finney. I also seem to remember from way back when a statistical book by Snedecor and Cochrane (spelling?) and a statistical book by Steele and Torrie.
The best software for this is GraphPad Prism.
With a large number of nonlinear models
Of course, I also recommend the lovely Compusyn software.