[dalke]

> "In drug development however, it takes a very long time to know that a candidate for a drug doesn’t work."

Mmmm, only somewhat. At the beginning there are million or billions of candidates. Most of them are easily rejected. It's only a few of the candidates which get far enough along to go into animal or even human testing which cost the big bucks.

> "Tests in a lab can be automated, are a lot cheaper and take much less time to execute."

As in, combinatorial chemistry? Big arrays of screening robots? DNA-encoded chemical libraries?

> "I claim that we should turn our efforts to: 1) Make the in Vitro testing phase a more profound step in the process and as a result:"

Yup, sounds like combinatorial chemistry.

> "2) Find more tests which together provide better predictability of: a. More potency, b. Less toxicity"

Says pretty much everyone, for decades. It's hard to go to a QSAR conference and not hear about someone trying to do this.

> "3) Enable each iteration to make a small change to the candidate and repeat the process."

Yes, this is called QSAR.

> "Nowadays, if a drug fails the clinical trials phase a biochemical engineer may try to slightly alter it — and repeat the entire drug development process from scratch."

Ummm, what? No. It more often goes back to an earlier stage. It doesn't restart "from scratch." If you've got a good lead compound, you're going to "make a small change to the candidate", not go back to, say, virtual screening of millions of compounds from chemical space.

I don't think this author knows much about how drug development is done. Why is this linked-to from HN?

[osharav]

QSAR sounds very interesting, if I encounter a failure in later stages of drug development - can I feed that information back to the model? See I'm trying to understand if the concept of a unit test is present in the drug development field. Unit tests watch your back - whenever a bug is found you are required to both fix the bug and write a unit test which ensures the bug won't happen again. If QSAR has that - my article may well be redundant. The author, me, knows a bit about drug development. But I do know about software development. If you find this not useful, read on to the next article. > "Why is this linked-to from HN?" This is linked from HN because I've published it there. I guess you can contact the administrator to notify him/her that someone is wrong on the internet.

[dalke]

As with just about everything to do with chemistry and drug development, the answer is "maybe".

It depends on the model, it depends on the failure, it depends on so many different factors.

For example, it may be that in human trials the new drug X is better than nothing, but not as effective as an older, generic drug A, which costs $0.10 per pill.

That's a market failure, but it's not possible to feed that back into the model.

Or, look at the TGN1412, where all 6 volunteers for first-in-man Phase 1 tests has to be hospitalized; 4 with multiple organ disfunctions and may never fully recover.

That doesn't require a simple fix to the existing model, but a investigation into how the underlying mechanism works.

But in general, yes, absolutely the information is fed back into the development process.

That said, I can make no sense of your analogy to unit testing. Drug molecules aren't meaningfully decomposed into individual units that can be tested, and a modification in one structure which may be deleterious may be beneficial if done to another structure. There are rules of thumb ("Rule of 5", "magic methyl"), and more statistics-based model ("Free-Wilson model", "matched molecular pairs"), but they do not ensure that there will/won't be a problem, only enrich the likelihood of finding what you are looking for. Hopefully.

Based on what you've written, including your lack of knowledge of QSAR, no, I don't think you know much about drug development.

You have a new account, so I will explain what my comment means. At HN, the readers are also semi-moderators. By making my comment, I was doing the first step towards "contact[ing] the administrator", and letting the submitter (that is, you), know that it is a poor fit for HN.

The HN guidelines say that topics should be on "Anything that good hackers would find interesting." I am a good hacker. I also write software for drug development. I did not find the piece to be interesting. I found it to be dismissive of the hundreds of thousands of people who have worked in drug development, and long ago put iterative processes and quality control methods in place.

I guess if you don't like my comments, you too can file a complaint with the administrator. I will continue to criticize articles that I think have little understanding of topics that I know something about.

[osharav]

Fair enough, although I think that discussing and further thinking about that analogy might be able to create new insights. I find it unfortunate that in addition to highly interesting information you've shared, you have also chosen to make destructive criticism about my attempt to find touch points between software and drug development. See, software development, the way I see it, is based on sharing, via open source, but not only. The culture it induces is of being open, to new ideas and to constructive criticism.

> Drug molecules aren't meaningfully decomposed into individual units

It's not about the molecules - it's about the tests. Envision a state where you could say - "oh dear, 6 volunteers for first-in-man Phase 1 tests had to be hospitalized - let's find a lab test, or a change in the model which would predict a positive result for the structure we've tested and negative result for other structures which do not cause that condition". The individual units don't have to be the molecules, they can also be the tests. You would argue, and rightly so, that finding such a test is hard/impossible. And I would agree but the whole article is about defining the problem - and because in software development, when making a small bug fix - you are not after making the entire system work, rather a tiny fraction of it better apt for a new condition you did not take care of earlier.

> the answer is "maybe"

When uncertainty is high - wouldn't you want to strengthen the tools that allow you to know at least that the failures you've already encountered won't happen again?

I find it also unfortunate that you think my article is dismissive of efforts made by countless people before me. I had and have no such intention.

[dalke]

If you think my comments are "destructive criticism" then I invite you to post to a medchemist's site, like Derek Lowe's, or present at a medchemist's conference and see the responses you get.

By giving you all these pointers I am trying to provide the "constructive criticism" you say you want.

I am also trying to establish my bona fides, so you might have a chance of accepting as true my statement that you've offered no insights which were not already known a century ago. It's not like the concepts underlying unit testing didn't exist before CS developed them.

And I have no wish to put up with an eternal September here on HN when some enthusiastic engineer thinks the methods from that field can be moved over to drug development. It very similar to what Lowe, above, calls the 'Andy Grove fallacy' - http://blogs.sciencemag.org/pipeline/archives/2015/04/02/sil... .

Also, there's an entire field for the theory of the design of experiments: https://en.wikipedia.org/wiki/Design_of_experiments . These are used in drug development. Perhaps it's software development which should incorporate ideas drug development, and not the other way around.

[osharav]

I am not saying that Computer Engineering (not science) can magically overcome the challenges of drug development - that would indeed be fallacy. I am saying that one of the challenges in software development is to maintain progress in an ever changing world of modifications done to the product. I see similar challenges in the drug development field - And so I try to find the link between what has worked in software development to what exists in the drug development area. You say that due to the complexity of drug development and the huge difference between the two fields all I can say is that in both worlds we conduct experiments and in the drug development area you have a far better understanding of how to conduct these experiments effectively.

You know better than me. You're right, I can offer no new insights, only the knowledge gained from experience (not necessarily theory). And from my experience what makes a product work is not the algorithms used, or the number of people working on it, or the technology used - it is how its progress is maintained via automated/constant testing.

If you think that the resources spent on maintaining progress in the drug development world are suitable - that's good to know.

[dalke]

Quoting from that Derek Lowe link to http://blogs.sciencemag.org/pipeline/archives/2015/04/02/sil... since it seems to describe your situation almost perfectly:

> There’s another problem that’s not unique to [Silicon] Valley, although it does tend to give people a bad case of it. That’s the “Clearly I’m smart and successful, so clearly I have something to offer in this other field over here” one. We all succumb to that one now and then; it’s human nature. ... If you’re used to being able to sit down and bang out code, any time, anywhere, with all kinds of tools (libraries, compilers, virtual machines, what have you) at your fingertips, then yeah, working up a new assay protocol in a cell line is going to seem agonizingly slow. Multibillion dollar ideas can be cranked out in the coding world very quickly, if you hit the right place at the right time, but just you try that in the lab. ... The real bottlenecks are figuring out what assay to run, and what to do with the data once you have it. ...

> As much as I might like to see something like that happening in biopharma, though, I can’t quite make myself believe it. Technology, Silicon Valley style technology, is human-designed and human-optimized for other humans. As human beings, we’re playing on our home turf there. But the biology of disease is an away game if there ever was one. The inner workings of cells and the ways that they work together are flat-out alien compared to anything we’ve ever built ourselves. People who are used to coding up apps have never experienced anything like it, and many of them don’t seem to realize that they haven’t. Expecting the sorts of behavior that you get from human-built technologies, and expecting the same effects from the techniques that work to optimize them, is an expensive accident waiting to happen.

As a related example, in the early 1990s AutoCAD thought they could enter molecular modelling, since they figured they could leverage what they knew from designing structures to design molecules. https://www.fourmilab.ch/autofile/www/chapter2_82.html . You'll notice the lack of success, and HyperChem is a dead product.