My gloomy mood this afternoon was greatly lifted by the headline provided by the folk at Nature Chemical Biology for a paper actually entitled “Global survey of cell death mechanisms reveals metabolic regulation of ferroptosis”. It certainly made me reconsider my decision to eat so many scones with clotted cream in the last week:
During our recent Journal Club, we discussed the paper “A real-World Perspective on Molecular Design” by researchers from Roche. The paper presents 10 case studies where computational methods were successfully applied to different research projects. The examples are diverse enough to cover a broad spectrum of possible approaches that one may take when doing computational study. It provides a set of guidelines, a check list of computational things you can do that are compatible with short project deadlines and that can provoke further thinking and generate new ideas. It also reminds the users that although computational methods, by their nature, provide numerical values, these should be treated as a qualitative input to an ongoing project. Somewhat surprisingly, all the studies were given in a prospective manner, which rarely is the case with published computational approaches – this in a way highlights the real life, practical use.
However, it is not clear who the article is intended for: for the method users/developers it lacks depth and for experimentalists it might be overly simplified, giving the impression that these are simple approaches that work every time and which can solve anything. But, as we well know, the road to success is paved with failure. It’s a shame these failures are rarely recorded in public literature. The authors are users/developers of the methods, so we find it surprising that they consider further development of the methods useless, “further improvements in computational methods may then have less to do with science than with good software engineering and interface design”. We think that science will benefit from improvements in computational methods and vice versa. The two are tightly interlinked: there is no good software engineering without the good underlying science. However, good science can exist without good software, but good software can surely help the science.
Unfortunately, they mentioned just cases in which protein structures and ligand properties are well known. Earlier stages of molecular design lack that sort of information and thus might need implementation of different approaches, such as computationally more expensive tools that haven’t been mentioned in this perspective.
We also found that some of the words are not always used appropriately. Looking at following sentence ‘Sulfonamides are molecular chimeras, which are found to form hydrogen bonds as well as interact with unipolar environments within proteins.‘ it seems that rather then being a hybrid chimeric molecule, sulfonamides as described in the article are more of a Dr Jekyll and Mr Hyde type of molecule which can be either one or the other rather than being a bit of both at the same time. Quite what the authors really meant by unipolar is also slightly confusing; if they mean a single magnetic or electric pole (unipolar), this seems unlikely to be a feature of many protein binding sites, at least when getting in to the interesting detail!
The last blog post on the subject of Roundheads and Cavaliers prompted me to reflect about how good I think the analogy is and whether it is useful. It eventually got me wondering about another, related subject that I have struggled to find the words to describe. This is the divergence between the mindsets of two sub-disciplines of chemistry: synthetic and medicinal chemistry. Although it may no longer show, my PhD training was in synthetic chemistry. One of the things that I enjoyed most during that period was the retrosynthesis challenge that would be set for the group. We would work in small groups to try and concoct a convincing synthetic approach to various fiendish molecules. These would then be presented to the rest of the group who would provide robust criticism of the proposed schemes – it was a terrifying undertaking for a first year graduate student but a satisfying thrill by the time I was finished. The great challenge was to have a synthesis that would get past the immense cumulative knowledge of a world-leading bunch of synthetic chemists. The whole session hinged on the transferability of reactions from one context to another. Those who knew the most detail from the literature would always “win” – and what is more, this was (almost always) because they could reasonably say that they knew what would work and what would not. Hence, their proposed syntheses could be better informed than everybody else’s and they could provide more informed critique of the other syntheses.
This is an appalling preparation for medicinal chemistry.
It has frequently shocked me to hear medicinal chemists pontificating about what will and will not work. It has too often felt like a delusion. Actually, that’s not quite right. In terms of stacking the odds in your favour, it is a pretty good idea to say that everything will not work. This makes medicinal chemistry ideal territory for self-satisfied Roundheads. But how unhelpful, how uninspiring. In an environment in which little is understood definitively, we require persistent sorts who can take the knocks of things not working as hoped and can pick themselves up and do it all again. A corrosive presence that will decrease the prospects of success is the one who can only tell you why they think something will not work or worse, the “told you so” sorts who don’t even make testable predictions. I fear that the puritanical roundheads of synthetic chemistry are all too often in this category. Is this really the best training for the random bittersweetness of drug discovery? What sort of training can we provide that will bring more joie de vivre to the undertaking, where do we find cavaliers and how do we train them?
OK, the title is probably not very accurate and I am sure that those more historically literate than myself will highlight why this is a bad analogy. Its also very UK-centric for which I apologise for indulging myself. However, I think it is a useful comparator to explain what I think is wrong with many of the bandwagons that go zooming by if you just wait around long enough in a drug discovery environment. Its not so much the roundheads as having a distinctive look or political grouping that is the source of the analogy I wish to draw but their puritan roots. In particular, it is the puritan tendency to tell people what they should NOT do and their propensity for banning things which is the basis of the comparison I want to make. Notably, they undid themselves by (amongst other things) banning things such as the various feasts and festivities and the vividly decorated public buildings that were some of the few sources of gaiety for many of the populace.
The comparator that I wish to make and which therefore, I think makes me a cavalier, is that there are too many people trying to tell drug discoverers what NOT to do but without providing any particularly useful guide to what they should do instead. I further speculate that, like the roundheads, this approach is wont to sap the joy out of drug discovery for many and ultimately is unlikely to spur the kind of creativity that we cavaliers think essential for success in this field.
Compound related metrics The first puritanical approach that I wish to highlight is that of the various drug discovery metrics and whatnot that sometimes get restyled rules (eg Lipinski’s rules, the rule of 3, etc). Other such metrics include ligand efficiency and lipophilic efficiency and the myriad related functions. My former manager, Pete Kenny, has critiqued many of these from a scientific perspective and others have challenged their rigour while their supporters have argued for their utility as well as their rigour. I should acknowledge that I have personally found lipophilic efficiency a useful guide for contextualizing compounds relative to one another and to be a thought-provoking concept. However, it is a very poor tool for suggesting what to do next. It is a much better tool for rapping me across the knuckles for daring to think about suggesting more lipophilic compounds.
Druggability (protein related metrics) In my research group, we have most enjoyable group meetings on a Friday afternoon at which, from time-to-time, we discuss journal articles. This week’s is a perspective by Kozakov et al. describing druggability. I was once more struck by the roundhead tendency of the approach. Targets can of course be undruggable but deluding ourselves into thinking that our understanding of biology and chemistry is so complete that we can predict this in advance using only the structure of the protein involved is not just puritanical but rather presumptuous too. The authors do not help themselves by mentioning that some druggability approaches consider HMGCoA-reductase to be undruggable. My main problem with this concept is that it tells you only what NOT to do and while I appreciate that drug discovery is an enterprise in which the resources available are unlikely ever to be great enough to take on all possible targets and approaches, this seems to allow no role for curiosity and the human delight in exploration. Indeed, if this is really a tactic for wrapping up a decision about resourcing as science then we really should avoid that – a resourcing decision is a resourcing decision and should not be disguised as something else.
Forbidden substructures I am aware that so far, you could read this article as me trying to tell you not to do things that tell you what not to do. To try and provide an illustration of what I hope is a more constructive (the word cavalier in this context might make me sound rash) approach to tackling roundheadism. I am sure many people working in drug discovery have come across those who would ban certain substructures, sometimes with good reason, but mostly based on extrapolating one or two bad actors to a whole class. I have heard tell of the banning of nitro groups and, of course, of aromatic amines. I was particularly vexed by the latter because I had been led to believe that the problem with aromatic amines is mostly a chemical one: they are transformed biochemically into reactive species but then undergo a purely chemical reaction with genetic material. I thought I understood chemistry and so presumed that the problem could be tackled logically. I think we demonstrated in a couple of examples in real drug discovery projects that this is correct and that you can find examples that retain all the “good” properties but which are significantly less risky from a DNA reactivity perspective (I put it no more strongly than that). I would far rather hear about interesting new ways to make logical (or illogical) progress in drug discovery than to hear new ways of telling people what not to do.
I think I will leave it at those three examples for now but expect to feel compelled to rant about further roundheadism in the not too distant future.
Iva recently attended the London Innovation Society’s Big Data Analysis Innovation Awards at which she was selected to present a poster. This has prompted us to ponder whether our recent collaborative work with MedChemica is genuinely “Big Data” or just an analysis that happens to have more data than is normal in the field (Medicinal Chemistry). Moreover, what (if anything) can we learn from the leaps forward in big data analysis taking place in other sectors? A recently published article considers how genomics might compare with astronomy, YouTube and Twitter (if nothing else, we enjoy the juxtaposition of one of mankind’s most primordial obsessions with the obsessions into which we are now regressing). In terms of sheer scale, medicinal chemistry seems to still be some way off from having the “zetta” (10 to the power of 21) scale data attributed to genomics or astronomy. Depending on how inclusive one wishes to be, it may compare with the fractions of a billion tweets per year. My back-of-the-envelope guess is that the global medicinal chemistry effort might add some hundreds of millions of datapoints per year (an HTS may be of order 1 million but few organisations can undertake them; individual compound testing efforts within large companies may add hundreds or thousands of data points per active research project of which there may be some hundreds). Recent efforts to make and test encoded libraries with billions of compounds in them probably don’t yet add one data point per compound so are unlikely to shift this in the near term. Indeed, it is not clear whether the number of medicinal chemistry data points being generated per year is currently increasing or decreasing. It is a thought provoking contrast that the four-headed beast that is predicted for genomics (data acquisition, storage, distribution and analysis) remains barely relevant to medicinal chemistry: the data instead remain divided amongst a stack of individual companies around the globe. Databases like Chembl (13 million datapoints) surely represent only a small fraction of the medicinal chemistry dataset but not an insignificant fraction. Others and others have recently speculated about the impact big data will have on medicinal chemists. Two aspects that we are particularly interested in are training and culture.
Unless things have changed radically in the last five years, most medicinal chemists come from a background in synthetic organic chemistry. As has been noted recently, this is the discipline of the discrete, the precise and of worrying about how to make things. Medicinal chemistry on the other hand deals with the “continuous” properties of biology which can be measured with much less precision and reproducibility and should be concerned with what to make (how to make it only kicks in afterwards). Does this training provide the best background to deal with medicinal chemistry in the big data era? What role is there for statisticians, analysts and mathematicians? Particularly in the UK, can we start to bring back some of the brilliant minds that have been lured into the city to do just this sort of analysis? Furthermore, a culture that is imbued with the beauty of synthesis (a penchant that I still share, but these days more as a guilty indulgence than anything else) and on caring for individual molecules (I have chosen the verb “to care” with some care: “the process of protecting someone or something and providing what that person or thing needs” describes one aspect of the problem rather well). It is hard not to have your head turned by any molecule (or other thing) that you have invested many days of hard work to but making the right thing may require just that.
The Leach research group currently consists of Andrew Leach, Iva Lukac and Joanna Zarnecka. We intend to use this blog to keep track of what we are up to and share some of the cool things that we discover along the way!
Welcome to the Leach Research Group blog. We are based in the School of Pharmacy and Biomolecular Sciences at Liverpool John Moores University.