ChEMBL

I am the world's best procrastinator, always putting things off - today I've actually got a whole bunch of stuff done - the motivator for this was promising myself that once I'd done a big pile of stuff, I'd write something for the blog. Yes, I'm weird, but I find that writing this stream os ASCII characters on your screen right now, makes me think about the way to explain things to others, and also makes me ask the important question 'is this likely to be of any interest to anyone?'.

Anyway, here's a toy analysis - Clinical attrition rates are the subject of a lot of analyses, the most rigorous ones usually involve the survey of a set of large pharma companies, and then analysis of the attrition. This is simple, and has led to various estimates of around 1:10 survival from phase 1 to launch to a 1:20 number. So 90 to 95% of compounds entering clinical trials fail, success probability is about 0.05. My hunch is that this is an underestimate - Logically, large pharma should be the best at pushing things through development, and if analysis is restricted to this best-in-class subset, surely the attrition of the remainder should be higher. Let's do a back of the envelope calculation, and let the data speak - or at least point us to the data we actually need to put together in order to do it right. I don't have time to point to some references for these numbers, that is for another day; remember, I'm only half way through my list of urgent things to do, or I will be in trouble.....

So, picking up on the discussion of kinase inhibitor attrition recently - about 20% of the USAN named compounds make it to market - this is roughly the same as the non-kinase set of drugs (data not shown). So this is 1:5 attrition (or 0.2 probability). Now, for the kinases, we have a pretty good set of data for all compounds that have entered clinical trials, and the ratio of non-USAN to USAN compounds is about 3:1, so using a steady state assumption (which is probably reasonable, data not shown, but the subject of a future post), gives a transition probability from entering clinical trials to non-proprietary name assignment of 0.25.

So overall, the attrition, from this small, but well characterised set is (0.2 * 0.25) which is 0.05, or 1 in 20. So things hang together quite well, and the number is sorta consistent with other analyses. Next some correlation with chemical properties.