Why is the best-fit value of the EC₅₀ very different than what I expected?

The usual cause for this problem is misunderstanding the definitions of an EC50.

The EC50 is the dose of a drug that provokes a response half-way between the top plateau and bottom plateau of the dose response curve. Unless you constrain the Top and Bottom to constant values based on controls, Prism will do its best to find the best-fit values for the Top and Bottom plateaus. If your data don't really define clear plateaus, it is quite possible for the best-fit value of the top plateau to be much larger than your largest data point. And it is quite possible for the best-fit value of the bottom plateau to be way below your lowest data point, and maybe even negative. If the Top and Bottom plateaus are not well defined by your data (or constraints) then the EC50 is not well defined either.

In many cases, you can fix this problem quite simply. Constrain the Top plateau to a constant value based on control. And constrain the Bottom plateau to a constant value. With these two parameters defined as constants, the EC50 is the concentration that provokes a response half-way between them. Of course, this approach is only helpful when you have good quality control data to use to define Top and/or Bottom.

Note that the EC50 is NOT the concentration that provokes a response Y that equals 50.