Skip to Content

The Drug Project Landscape

Here’s a new paper in Nature Reviews Drug Discovery that’s going to the trouble of matching specific disease indications with specific mechanisms in drug projects over the last 20 years (both the successful ones and the unsuccessful ones). The authors (from Vertex) used the Cortellis commercial database (with a good deal of filtering and manual curation) to assemble this list – as far as I can see, it’s probably the most comprehensive thing of its kind. The total set included about 10,000 unique drugs, representing over 22,000 reported projects. That boils down to about 2,400 mechanisms for 1,400 indications, and the combination of those two yields about 15,000 mechanism/indication pairs (things like “calcium channel blockers for angina” or “DNA gyrase inhibitors for Streptococcus infection”).

If you break things down more broadly by disease, the conditions that have had the widest variety of therapeutic mechanisms tried against them include breast cancer (and many others, like pancreatic and non-small-cell lung cancer), rheumatoid arthritis, Alzheimer’s (of course), pain, Type II diabetes, multiple sclerosis, and Parkinson’s. No one who’s been around drug research will be surprised one bit by this list, but it’s good to have it quantified.

But looking at the more granular data set is interesting, and it’s interesting at both ends. Those two pairs I mentioned above are, as you’ll have surmised, from the more successful (and more crowded) end. There are 128 mechanism/indication pairs with more than 5 successful projects (and these alone account for 22% of all the successful projects in the data set!) To be sure, some of these are classic “me-too” projects that came along years later, but many of them are the (equally classic) “footrace” situations, where several companies were pushing into an area more or less simultaneously and fighting it out. As the authors put it:

Regardless, this fraction indicates that a significant proportion of the pharmaceutical industry’s output in terms of new drugs (and probably also revenues) in the period studied has come from a small fraction of repeatedly validated pairs, indicating the challenge of identifying truly attractive pairs for which scientific possibility overlaps with clinical need and commercial potential.

Yes indeed. You sometimes have to squint to see the overlap in that particular Venn diagram! But keep these figures in mind: of the projects that pursued an already-validated pair, 37% of them were discontinued (this in the most favorable possible situation). And 11% of the total mechanism/indication pairs actually have more discontinued projects associated with them than successful ones. There are bleached bones everywhere you look, basically, even in the most successful and validated parts of the landscape.

If you look at the pairs with the highest number of successful projects and no failed projects, you find things like proton/potassium ATPas inhibitors (omeprazole and such), H1 histamine antagonists (the familiar antihistamines) and other widely recognized winners. But on the other end of the data set, there are mechanism/indication pairs that have never yielded a drug, despite many attempts. Among the pairs with no current projects (thus probably totally abandoned) there are, for example, 28 known efforts to develop ACAT inhibitors for atherosclerosis. None worked. The folks in the lab next door to me, when I started in this business in 1989, were working on an ACAT inhibitor, come to think of it. Potassium channel stimulation for hypertension is up there in this category, as are NMDA antagonists for cerebrovascular disease and PAF receptor antagonists for asthma.

In the “unvalidated but people are still trying” category, the most striking thing is the horrifying domination of Alzheimer’s projects. If you distinguish between amyloid synthesis, amyloid deposition, and amyloid modulation (as this data set does), you end up with ninety-three known discontinuations. Even backing off to the pairs where something has worked, you can find horror stories. 5-LOX inhibitors for asthma, for example, are 1 for 30, and NS5B polymerase inhibitors for HCV have yielded two successes out of 49 known attempts. Other notorious rocks like Factor IIa antagonists for thrombosis and renin inhibitors for hypertension show up on the list – the amount of time, effort and money spent in this categories is fearsome to imagine.

But these represent the last twenty years of reported work – what about the current situation? The authors take a crack at that question, and conclude that things have been changing. There are 1241 ongoing projects in their database. Only about 8% of them are targeting mechanisms that have been repeatedly validated, and another 18% are targeting something that’s been validated once (a total of 231 projects working on something known). That leave 1,010 projects going after unvalidated ideas. 30% are unvalidated with at least one failure on the record, and 70% of those are going after combinations that are so new that there have been neither successes nor failures for them yet. These are definitely above the historical averages – see above.

So there are some hard numbers for the “All the drug industry does is work on stuff that’s already known” criticism. It isn’t true. We’re out here on the limb, folks, we’re out here on a lot of limbs, and a some of us are going to end up landing hard. That’s research.

11 comments on “The Drug Project Landscape”

  1. bhip says:

    If you were willing to lump all indications into a single “chemokine receptor” bucket, I suspect that the carnage statistic would look be impressive….

  2. Anon says:

    “5-LOX inhibitors for asthma, for example, are 1 for 30, and NS5B polymerase inhibitors for HCV have yielded two successes out of 49 known attempts.“

    That’s even lower than the odds of getting a statistically “significant” result with a p-value of less than 0.05 based on random sampling of the null hypothesis. Makes you wonder how many approved drugs really do work???

    1. Anon says:

      We should keep trying for amyloid agents against AD, because we’re bound to get a statistically significant result with one of those sooner or later, even if it’s a crock of crap. In fact we’ve just been darned unlucky at 1 success in 93 attempts. Reminds of green jelly beans!

  3. Petros says:

    5-LOX inhibitors for asthma, for example, are 1 for 30

    And that one, zileuton, was launched as a 800 mg q,i,d, drug!

    I remember the internal efforts on compounds that acted on 5-LOX via FLAP and it was an area where many major pharma where active, then some small companies tried to exploit the failings

  4. Calvin says:

    However, for NS5B the success did lead to Sovaldi which made quite a bit of money (must be well over 20B by now). And some of the “failures” in NS5B were projects that ended once it became obvious that Solvaldi was going to clean up, so not technical failures. So I think some care need to be used here when looking at the data.

    1. anon says:

      Yeah, one wonders how many terminated projects are due to being beaten to market (or other non-pharmacologic reasons, like management decisions), as opposed to someone recognizing “guys, this useless molecule belongs on the trash heap, end of project”.

      1. exGlaxoid says:

        I would guess that half of the discovery projects I worked on failed due to cancellation my marketing, mergers, reorgs, political issues, competitor products, or other issues rather than lack of scientific success. We had at least a dozen compounds that looked very promising discontinued after being declared a lead optimization candidate which appeared to meet all needed goals. The number was lower in development, but I do think a few drugs which meet their clinical endpoints were cancelled due to reasons other than lack of success.

  5. Dr CNS says:

    So, do we still think bringing a new drug to market costs only around $2 billion?

  6. Martin S says:

    This paper is very interesting, but I wonder about the complete accuracy of the analysis. For example, in the heat map of figure 3c, the AMPA receptor antagonist section has no entry for epilepsy syndrome. Perampanel was approved for this indication and is a non-competitive inhibitor of AMPA (should be blue). Could there be other missing or inaccurate items?

    1. Alex A says:

      Perampanel was first approved as an adjunctive therapy, and later as mono therapy, for partial seizure with or without secondarily generalized seizures. In Cortellis it is annotated for partial seizures rather than the broader indication, epilepsy. Thus, this drug and its MOA are part of the dataset, not missing.
      Regardless of this specific example, indication ontology is an issue in the field, and something that was discussed at length in the paper.

Leave a Reply to Calvin Cancel reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.