A leading industrial drug discovery leader was significantly using molecular modeling to engineer selectivity for a tough kinase for 3 whole years. When I asked him how much of a role modeling played in the ultimate success of the program, his response was "None"

Now the usual response to this would be "If they had used [favorite method X] I bet they would have been successful." A worse response is "They weren't good modelers." The best response is "What would be a good system to record those negative results and learn from the failures?"

One of the most important messages here (and I congratulate this person for being so honest) is that if a tool you have invested a lot of time and money in doesn't work, say so. Otherwise you will create an illusion of success, as happens often especially with computation.

I've always said that if you get together a group of computational chemists working in drug discovery and ply them with snacks and drinks, half of them would admit their techniques aren't too effective. The other half would say their techniques work but won't be able to prove it.