Machines and MedChem

It has been interesting reading Derek Lowe’s articles about machine planned synthesis and the pharma research space.

Much of this sounds like what software went through as compilers came on the scene and became progressively better. For a long time people would argue that human generated machine code would always be better than compiler generated code and that compilers were essentially a crutch for those unwilling to do the hard stuff.

Look around today and we see where the tools have brought us. I’ve observed in the past that the duration of the average project doesn’t seem to have changed all that much. What has changed is the scope of the result. We build complex graphical interfaces with extensive business logic behind them in the time that engineers fifty years ago would have built a simple, command line driven program with a much less complicated back-end.

I expect that what we’ll see in chemistry is something similar. The chemists will stop worrying about how to attach a particular functional group or what the path to their desired end product is. They’ll instead ask their machines about best case yields and production feasibility and then let the software systems optimize the synthesis paths. I can imagine the equivalent of debug and release builds here as well…one approach that works well for small batches using laboratory equipment and perhaps inputs that are expensive, but get to the end-point more quickly and another approach that is tuned for volume production from the least expensive feed stocks that make sense.

I do expect that the scientists making the transition will find it stressful. I expect that some will not be willing to make the change. I think that in the end, those who remain will be happier and more productive using the new, powerful tools that come out of the machine learning/AI labs.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.