This is not a paper that’s going to make everyone who reads it happy, but it needs to be read anyway. A collaboration between the University of Helsinki, LifeArc (which looks to be one of the small companies in the former Stevenage pharma campus) and Cyclofluidic reports development of inhibitors against hepsin, a serine protease enzyme that is a potential cancer target.
So far, so uncontroversial. The same group had identified a micromolar hepsin inhibitor from some late-1990s urokinase inhibitor chemical matter, but this compound (naturally) still had activity against urokinase. The goal was to improve potency while working away from urokinase, which is just the sort of thing that med-chem teams everywhere spend their efforts on. The enzyme seems to need an amidine/guanadine functional group for potency, and a screen of commercially available derivatives gave another micromolar hit with less urokinase activity as a new starting point. 142 new compounds were prepared, optimizing along until a final compound was reached with 22 nM potency and 200x more selectivity in the counterscreen.
Which is good: that’s what a team of medicinal chemists are there to do. What makes this paper interesting (and those who have heard of Cyclofluidic will have already caught on) is that the chemistry was automated in a flow reactor, and the enzyme assay results were fed back into the software which generated SAR predictions for the next round of synthesis. Importantly, the screening assay is part of the automated system as well, and the entire loop (synthesis/purification/assay) is claimed to take around 90 minutes.
The way this actually worked was that the automated system would run through a particular SAR space, generating a series of compounds, and at that point the humans involved would have a look at how things were going. Subsequent runs might include new SAR building blocks or changes to the algorithms that evaluated the results. So it’s not like they walked away for a weekend and came back 142 compounds later to a selective nanomolar inhibitor – but neither is it the case that people made all these compounds and selected the next ones to make, either. This was machine-assisted compound development, very much analogous to what Gary Kasparov has termed “centaur” chess, where humans evaluate their ideas with the aid of software to come up with even stronger moves.
It needs to be noted that the chemistry involved was very straightforward (amide and sulfonamide formation, for the most part) and the assays were quite robust. This was close (as have been the other things reported in this vein) to a best-case example for such closed-loop automated SAR work, but then again these are early days. I also noticed that several of the co-authors on the paper now have different addresses, which makes one wonder how long ago this project was actually done. My point is that automated synthesis (thanks to human ingenuity) is getting better all the time, as are the methods for in-line assays and especially those for evaluation of chemical space and SAR. Thanks to human ingenuity, we are very likely going to be freeing up time to apply that ingenuity to other things than banging out obvious SAR compounds.
If, though, you are a human chemist (as so many of my readers are, to steal a joke from G. K. Chesterton), and you feel that you are watching your livelihood being stolen right in front of your eyes, consider what in this case the machines are taking from you. You are being deprived of the chance to do some of the most boring chemistry in the world in the most routine fashion imaginable. My own feelings are that any machine that takes away from me the necessity to crank out 142 amide reactions by hand is welcome to it. This is the med-chem equivalent of doing the laundry, and I have a machine for that, too.