Finding small molecule drugs is much harder than finding a needle in a haystack – discovering the right arrangement of atoms to bind precisely to a protein target to elicit a particular response is a problem of vast dimensionality.
We are most familiar with the numbers involved when dealing with antibodies: a typical antibody library might contain 1013 different clones – but even that hardly scratches the surface of the 1080 or so possible CDR sequences. Choosing the sequences to populate a library is critical – and even today new innovations, such as the RxBiologics’ Galaxy™ platform, are dramatically improving the output.
Yet the situation with small molecules is even worse. Chemists look on jealously at a universe of a mere 1080 possibilities for antibody CDRs!
So DrugBaron asked Nigel Ramsden, who heads up the RxChemistry team, whether the solution lies in bigger libraries and yet higher-throughput screening?
“In a word, no! Bigger won’t cut it – even with a 10-fold increase in library size, you are still testing only the most miniscule fraction of all possible compounds. We have to face facts – its NEVER going to be practical to test a material fraction of the universe of possibilities when it comes to small molecules
So the answer has to lie in quality rather than quantity. Which molecules to test, not how many.
In exactly, the same way Galaxy™ advanced antibody discovery by being smart about which clones to include in the starting library, we have to be smarter in choosing the molecules we test.”
So scaling up real, wet experimentation cannot improve efficiency in small molecule discovery. But what about in silico searches – are they the answer to the fundamentaly dimensionality problem? David Fox, Associate Professor of Chemistry at Warwick University and head of the RxChemistry medicinal chemistry team, thinks its only a small part of the solution:
“We can scale in silico searches well beyond the practical limits of any physical high throughput screen. But even processor time is not infinite and free, so searching ‘everything’ remains well outside the realm of possibilities
And on top of that, in silico searches (like physical screening) are imperfect filters – there are lots of false positives. So you would need to make many of the in silico hits to find one with the desired profile in a real experiment. Given that, there is no point including compounds in in silico libraries that are difficult or impossible to make
Instead, we need to increase the diversity of the compounds we do screen (whether in silico or for real). Right now, both virtual and real libraries are dominated by particular skeletons that are easy to build – and a lot of those are flat. The future is definitely 3D”
Ramsden expands on the new focus on ‘what’ rather than ‘how many’ molecules: “You can make libraries a bit bigger, you can increase efficiency of screening (for example with DNA-encoded libraries that make searching for small molecule hits even more like panning for antibodies) – but unless a good hit is present in the starting mix, you still won’t get anywhere
Potentially, you can make much bigger gains by improving the quality of what we put in the library than by modestly increasing the size or the screening efficiency. Certainly at RxChemistry, we are putting a lot of effort into improving the quality of what we screen – and learning lessons from our friends at RxBiologics. Watch this space.”
But improving the quality of the feedstock is only half the story with next-generation antibody screening platforms such as Galaxy™ though. The other part is the search strategy – they very cleverly mimic the way the human immune system searches for a high affinity binder, by evolving the library during the search.
“We are never going to be able to copy that with small molecules” opines Fox, shaking his head sadly. “But it does highlight something else that we do differently to most other medicinal chemistry groups – we have re-imagined the classical search strategy that takes us from the initial hit through to a lead, subsequent lead optimisation and candidate selection. That’s the part of the process that’s analgous to the library evolution phase of Galaxy™
The classical approach beloved of large pharma companies is really a brute-force method: make as many molecules as you can to assemble the SAR. A lot of resources are therefore poured into synthetic chemistry, and if the compound family is at all synthetically challenging it can require dozens, even hundreds, of chemists to churn out every imaginable variant.
Most of these molecules however turn out to be rather poor in one way or another when they are tested – often in ways that could have easily been predicted from the burgeoning SAR. But because the synthetic effort is the engine of the program, much of that effort is wasted because the thinking came AFTER the compound generation.
We have re-imagined the whole medicinal chemistry strategy. Our goal is to find the optimal molecule by making as FEW compounds as possible!”
Ramsden adds “Absolutely. You can find the answer by making and testing every variant you can think of. Or you can carefully select the variants that are likely to answer specific questions. A good analogy are key frames in animation: you don’t need to manually generate all the ‘in-between’ frames to deliver smooth motion – it is defined completely by the beginning and the end positions.”
But in order to make such a change in medicinal chemistry strategy, you have to abandon the normal metrics of med chem progress, and even the way you structure your program. Instead of counting FTE chemists and numbers of new compound registrations each week (metrics that just incentivise activity over productivity), you need to look at decision quality.
“Experienced medicinal chemists exposed to our strategy for the first time are often shocked how little synthesis capacity we need to get to a clinical candidate. We don’t just make molecules, we make the right molecules – saving our clients money, but also time.” says Ramsden proudly. And their track record eloquently backs up his assertion.
“We call it the ‘tyranny of the n+1 molecule’” adds Fox. “The idea that making more molecules is the fastest way to answer the most questions. But random walks are rarely the quickest way from A to Z.”
This is actually true in any complex system, and DrugBaron showed a decade ago now that in low-validity systems (where the ability to predict the outcome from existing data is low), that the incremental value of each additional datapoint is not worth the cost of acquiring it. A whole range of Monte Carlo simulations suggest that making decisions based on less data is the quickest and most efficient way to proceed
And there is no reason to assume that medicinal chemistry strategy for finding small molecule drug candidates would be any different
“But it goes even deeper than that” interjects Ramsden “It’s about what KIND of information you want to generate. We already talked about choosing carefully a more limited set of compounds to make. But what you do with them is just as important. We would advocate spending more of the budget on understanding the profile of the compound you do make, and less of it churning out near-clones.
Another common pitfall in medicinal chemistry is optimise one characteristic at once: find super-potent molecules, then try and ‘fix’ the pharmacokinetics and then try and improve the pharmaceutical properties and so on. Yet it quickly comes to resemble a game of whack-a-mole – as you improve one parameter you suffer unacceptable declines in another.
By acquiring a much more data-rich profile of a small number of more diverse molecules, you can select starting points that nearer to optimal across all the parameters at once.”
Fox again: “That goes right to the heart of why, as a medicinal chemist, I work with RxCelerate. You have a platform that is as strong in biology as in chemistry, allowing close integration of the widest array of assays with the med chem strategy. Quick turn around of high data density profiling is the driving force for discovery productivity
What you need are answers to well-thought-out questions. Not just more data per week. Definitely not just more compounds per week. More answers.
It never ceases to amaze me how many molecules we have discovered against a vast array of different targets, while making such small numbers of compounds by following this ‘quality rather than quantity’ mantra, and avoiding the tyranny of the n+1 approach. But at the same time it amazes me even more how few people out there have grasped the benefits and are prepared to discard metrics of new compound registrations, and armies of FTE synthetic chemists, in favour of a process that is manifestly cheaper, quicker and better.”
I guess that little secret will remain a competitive advantage of those who “get it” for a little while longer yet!
This post was based on a conversation between DrugBaron, Nigel Ramsden (EVP, RxChemistry) and David Fox (Head of Medicinal Chemistry, RxChemistry) at Babraham, UK, on Wednesday 2nd February 2022. A video of the whole discussion is also available on the RxCelerate Youtube channel.
RxCelerate Ltd is an outsourced drug development platform based near Cambridge, UK. We specialize in delivering an entire road map of drug development services from discovery and medicinal chemistry through to formal preclinical development and clinical up to Phase IIa. In the last five years, we have witnessed dramatic changes in the drug development …
