Your description of the problem may be incomplete or somewhat unclear to me, but let me take a stab at it. It may stimulate something in your thought process to get the correct answer since you have more info than you've presented here (e.g., the data points).
Usually an inflection point occurs where your regression curve changes character, from concave to convex or the other way around.
Here your regression curve (or best fit to the data) is a straight line with a very high R2 of 0.994 implying that the line explains almost all the variation in the data. It's not clear to me whether this was before or after removing the two points?
If before (with all data present that was your R2), then there is no inflection point in a straight line since it doesn't change its character. For example, imagine a regression curve y = x3 which opens upward for positive x and downward for negative x. It clearly changes character and has an inflection point at x = 0.
If after, then maybe the 2 data points you removed represented a "new domain" that had a kink and a different regression line? I still would hesitate because 2 points determine a line with no variation around it so it is scant data to base a conclusion on.
My Comment Later:
Judy, This is my third try. This WyzAnt software is crap! Twice, I tried to add my comment to my original answer. I typed it all in, save commented it and it all got lost. Now I'll try a comment to your comment, and it will probably work. The software should either not have allowed the comment option on my answer that I entered earlier to type it in, or given an error message on the attempt to save comment so I could have clip boarded my text to place it here. Now I'll copy it before I conclude.
On to your question. You're basically correct. We can use the "kink" approach of two lines crossing each other at the inflection point. I get y = -1.019 x + 0.307 for the regression line based on your first 3 data points, and the line segment y = -0.06386 x - 3.4281 between data point 4 & 5.
The crossover point (when the y's match) is x = + 3.26 and y = - 3.02.
Also, if you look at the slopes of the successive line segments connecting adjacent points, they start out at -1.167, keep getting less negative and finally approach zero by data point 4 & 5. Thus, it looks like the drug is stable for x greater than the inflection point.