Once again, the problem is not synthesis, it's design

University of Illinois chemist Martin Burke who recently got a lot of press for his automated robotic molecular synthesizer has an interview in C&EN in which he says that his and similar other techniques will help to clear the bottleneck of synthesis that has plagued the pharmaceutical and other industries.

Chemjobber who linked to the piece disagrees and says, “I don't think synthesis has been a bottleneck for access to materials. It's cost of synthesis that is a barrier, or design that is the slow step.”

And I think he’s spot on. As I enumerated in some detail in a previous post, the defining molecular challenge of our time is not synthesis but design. I would go a step further and say that it’s not the design of structures per se but the design of properties – a much harder goal. Synthesis has now turned into a well-oiled machine which can now deliver the goods more or less on demand; the only real bottleneck is money and manpower. The medicinal chemists in my company for instance have told me that if I can sketch it, they can make it, or if they can’t make it, WuXi can. And I have found that to be generally true with few exceptions. Technologies like those that are being pioneered by Burke will certainly speed up the delivery of new molecules for applications like pharmaceutical chemistry and materials science, but I don’t believe they will cause a fundamental paradigm shift in the overall process of drug discovery and development.

I noted above that medicinal chemists can make pretty much anything you ask them to, with a few exceptions. The exceptions that I am pointing out don’t really relate to the impossibility of making specific molecules; rather they refer to tradeoffs. When a medicinal chemist tells me he or she is reluctant to make a molecule that I recommended, what they’re really saying is that the synthesis is non-trivial. But what they really mean is that they will make it only if I can justify the effort required to make it with a high degree of confidence. In other words, if I can actually predict the important and desirable properties that a particular molecule is supposed to have, then the chemist will be happy to put in the extra effort needed.

In drug discovery those properties can range from simple increases in binding affinity to ‘higher-order’ properties like toxicity, liver metabolism or clearance rate. In materials science they could be tensile strength, rate of hydrogen absorption or rate of solar energy capture. In drug discovery, predicting the binding affinity of a particular molecule is already a challenging goal which we however are now getting a handle on computationally, but predicting those higher-order properties is still a dim and itinerant light on the horizon. The problem is not in synthesizing molecules which will have these favorable properties; the problem is in having enough confidence in designing those molecules. What the medicinal chemists are telling me is that if I can solve the problem of design and prediction of higher-order properties, they have already solved the problem of making molecules with those properties.

That is precisely why, in a sweeping, highly readable 2013 essay speculating on the role that chemistry should take on in the future of science and technology, George Whitesides emphasized that chemists should move “beyond the molecule”. What he was referring to was again the design of goals and properties. His point was similar to the one made by Chemjobber: we are now really good at making molecules. A hypothetical, fully automated molecular synthesizer where you literally feed it a Chemdraw file and it spits out compounds in vials at the end would be a huge advance in speed and efficiency, but it’s still not going to automatically address the problem of design. And that's the problem which should really keep us awake at night.

1 comment:

  1. Same Phil as from Derek's post, with a little more time to refine my thinking. Other than making and testing compounds that are different from those we have already made and tested, what path do you see toward better prediction of higher order properties?


Markup Key:
- <b>bold</b> = bold
- <i>italic</i> = italic
- <a href="http://www.fieldofscience.com/">FoS</a> = FoS