A colleague and I were talking the other day about some of the molecules that turn up when you dig through a company's internal database. This was a favorite sport of mine during slow afternoons at the Wonder Drug Factory - I would put in a query for bizarre or unlikely chemical groups and see what fell out. I was rarely disappointed - eventually I assembled a folder of the most hideous examples, which never failed to astound.
The compound collection at my current employer isn't nearly so weird, fortunately. But every drug company has large lists of compounds that aren't so attractive as leads, because they were made in the last stages of previous projects. This is a well-known problem, often referred to as a gap between "drug-likeness" and "lead-likeness". For the most part, the compounds you start a project with don't get smaller - they get bigger, as people hang more things off of them to get more potency, selectivity, or what have you. So you're better off starting as small as you feasibly can, giving you room for this to occur without taking you off into the territory of too-huge-to-ever-work. (That's one of the fundamental ideas behind the vogue for fragment-based drug discovery, for example).
"Too-huge-to-work" is a real category, as my industry readers will gladly verify. I think that the "Rule of Five" cutoffs have been sometimes applied a little too mindlessly, but there's no getting around the fact that if your latest molecule weighs 750 and has thirteen nitrogen atoms in it, the odds of it being a drug are rather slim. As my colleague put it, when you make something like that and send it in for testing, what you're saying is "I know that almost every molecule that looks like this fails. But I'm different. I feel lucky". And that's no way to run a research program. Given finite time and finite money, you're better off prospecting in chemical areas with better chances.
So what to do? We kicked around the idea of setting up some filters in the compound registration system itself - if someone tries to send in some horrible battle cruiser of a molecule, the system would make a puking noise or something and refuse to register the compound at all. There would have to be be some sort of override (perhaps for a higher-level manager to authorize) for those times when you actually have evidence that the ugly molecule works, but maybe the "You Lose: Make Something Else" screen would focus attention on the properties of what's being made. Of course, if anyone ever implemented this, the arguing would begin about where to draw the line (maybe there'd be a yellow "warning zone" in between), but I think that everyone agrees that at some point a line should be drawn.
So, for my readers around the industry - do you have such a cutoff? Can you register any crazy compound that crosses your bench, or does your company's software fight back? If so, what's the feedback - beep, e-mail warning, electric shock? Inquiring minds want to know.