This was first posted: 04/03/2013 here
To be honest, I am not too fond of rules in general, and drug-likeness rules in particular. That’s just me. Or, perhaps, it’s not only me. In fact, there’s an increasing amount of evidence piling up disfavoring them. The idea of drug-likeness rules has been controversial ever since Lipinski introduced his rule of five for predicting oral bioavailability 15 years ago.
A personal example of an “unusual” compound (Elobixibat), in late stage clinical trials, violating common drug-likeness rules.
To clarify, drug-likeness is often defined by a set of very simple rules of calculated physicochemical properties and is intended to assist medicinal chemists to achieve their end goal; designing drugs. This is yet to be seen. On the contrary, many drugs do not fulfill the various rules. The first drug project I was involved in at AstraZeneca resulted in a compound (Elobixibat), currently in clinical testing. Elobixibat was designed before the drug-likeness concept (and metrics) engulfed medicinal chemistry thinking, and it does not adhere to too many of the rules.
There’s more to it; are the parameters used adequately described, for example what is an H-bond acceptor? Why use hard cut-offs (e.g. MW=499 is ok, but MW=501 is not)? How to treat interdependent x-variables (PSA vs. clogP?) Don’t differences in protein target classes (e.g. GPCRs vs NHRs) and pharmacology prevent the generalization of something as complex as drug-likeness? etc. This quickly adds up to the notion that drug-likeness rules are mere blunt instruments perhaps causing more damage than good.
There are other arguments against the drug-likeness rules. Having said that, some people state that these are not rules, but mere guidelines. I don’t buy that. They are most often presented as rules (and named “rule-of-N“) and are thus easily misused by management to measure compound “quality”.
A friend and a former AZ colleague of mine, Pete Kenny, recently showed that by binning noisy data strong correlations can be achieved (correlation inflation) casting a shadow on the validity on a number of drug-likeness studies. For example, a rigorous re-analysis of the “Escape from flatland” data set showed that what seemed to be a strong correlation (R2=0.97) turned out to be a very weak trend (R2=0.25) and of little value to the drug designer.
George Canning once said “I can prove anything by statistics except the truth”. That’s stretching it a bit too far. However there are limitations on how results are interpreted, the relationship between correlation and causation is one. That is, a correlation between two variables does not automatically imply that one causes the other. For example: “the more firemen fighting a fire, the bigger the damage is going to be. Hence firemen cause large fire damage, right?” Eh no. There are also legendary tales like "every time Wales win the rugby grand slam, a Pope dies, except for 1978 when Wales was really good, and two Popes died.”
Dr Payne of University Hospital Wales found weak statistical support (predictions were more or less correct because no popes died), but dismissed it as an urban myth. In a moment of despair, I made an (ironic) contribution myself: ”the number of approved drugs per year is inversely correlated to the number of J. Med. Chem. articles containing the word 'rule(s)'”. This highlights the common issue of misusing statistics – people tend to use and praise statistics when the analyses support their own favorite biases. And it is certainly not a huge challenge to find patterns which correlate with real-world trends over time. Just have a look at ”Google Correlate” for a few minutes.
To be fair, the era of drug-likeness rules was not all bad. Fifteen years on, we are all fully aware of that physico-chemical properties are important, and needs proper attention. Nevertheless, my personal take is that we can now sort them under the ”not even wrong” category and move on, quickly.
Finally, the emotional aspect is not to be forgotten. The mere existence of rules stifles innovation and creativity. It is rewarding to see medicinal chemists thinking in new directions, setting general drug-likeness rules aside, and focusing more on trends in structural series…and relying more on the unpredictable nature of creative invention.