Please make your sub-question the main question for more views.  Thank you.

Mohamed - 

You could try a graphical analysis of your estimated residuals to see what looks reasonable.

I am not familiar with your topic, but sometimes it depends upon the range of interest for x. 

In doing a graphical analysis of estimated residuals, you could be looking at a straight line in the range of interest, and separately, a smooth curve or two for comparison, but I generally would hesitate to use lowess, as you do not want to artificially play "connect the dots." 

Cheers - Jim 

Considering the sigmoidal shape of the dose-response curves, I think non linear model should be preferred over linear models.