[modeller_usage] homology modelling vs protein threading

Dear modellers,

I want to build models of ATP-binding domains with sequence identity ranging between 40% and 50%. In the FAQs it says that one needs 35-40% at least to build reliable models. However these models will be used for inhibitor design, therefore the accuracy of the structure prediction method will determine the outcome. Do you reckon the sequence similarity is enough to get a good model of the domain, or should I try other programs based on threading? Is there a way to compare the quality of the structures derived from these 2 different methodologies?

I would greatly appreciate any advice.

Thomas