When I was talking about chemical space the other day, I alluded to the attempts to cut it down to "druglike space" by use of rules of thumb. The most famous of these is Chris Lipinski's "Rule of Five", a summary of which can be found here. Lipinski and his Pfizer co-workers looked over a data set of drug candidates and noticed that there were some reasonably clear cutoffs for oral absorption and general cell permability. They suggested that you need:
1. Fewer than five hydrogen bond donors (which can be estimated by counting the total number of OH and NH groups in the molecule.)
2. A molecular weight of less than 500.
3. A logP of less than 5
4. And fewer than 10 hydrogen-bond acceptors (estimated by the total of N and O atoms in the molecule.)
The "rule of five" name came from the cutoffs all being multiples of five, in case you're wondering why there are only four rules.
(A quick explanation for non-chemistry types: "logP" is a measure of how greasy your compound is, and I mean that quite literally. Although it's often estimated computationally these days, experimentally it's determined by shaking a standard amount of a compound in a flask containing water and the oily 8-carbon alcohol, octanol. Those form two layers, as you'd imagine. The log of the ratio of how much of your compound goes into the octanol layer versus the water layer is your logP. A thousand-fold preference for octanol (a logP of 3) is considered to be just fine for a drug, so you can imagine how greasy a compound with a logP of 6 is.)
There's a recent paper by John Proudfoot at Boehringer Ingleheim (Bioorganic and Medicinal Chemistry Letter 15, 1087, for those of you playing along at home) which looks at a more comprehensive list of compounds than the original Lipinski batch. He finds that the cutoffs might be more like 470 for molecular weight and 3 for hydrogen-bond donors, but otherwise his analysis tracks Lipinski's pretty closely. (He notes that only a handful of drugs ever violate both those cutoffs simultaneously.)
His paper also includes a year-by-year analysis from 1937 to 1997. The only clear trend is that molecular weights have been increasing, from under 300 to the point where we're banging up right against that 500 line. Personally, the largest molecule I've ever submitted for testing weighed quite a bit more than that, but I had my reasons. It came in at exactly 747, and I couldn't resist.
Lipinski's done a lot of publishing and speaking on this topic in the eight years since he first published his rules, and they've seen a tremendous amount of use. Perhaps they've seen some abuse, too, because (as he himself would no doubt tell you), they're not written in stone. They're based on the 90th-percentile cutoffs for each property, so there are outliers. The "Rule of Five" is a tool meant to pare huge lists of compounds down to manageable size, not to blindly decide the fate of individual drug candidates.
I've seen people make Rule-of-Five objections to a compound after its in vivo behavior has already been demonstrated, to which the proper response is "Who cares?" You wouldn't want to stuff your drug discovery project with compounds that violate several of the rules, but if you have good activity after an oral dose, then what more are you looking for? Other things being equal, we try to make compounds that don't step over too many of these boundaries, but being dogmatic about them, especially single rule violations, is foolish.