Automated Compound Synthesis in Drug Discovery

Traditional early drug discovery processes are slow, and often the most costly part of the drug discovery process. I work in a group that aims to change that, by increasing the productivity of the chemist through automation. This is a hot topic in the industry right now, with massive companies like Eli Lilly and their infamous automation lab in La Jolla.

As an example, I recently have a colleague who joined our team as a medicinal chemist. She is a PhD with loads of academic experience. As somebody who spent a large portion of graduate school synthesizing various chemicals in prestigious chemistry labs, she has numerous publications. She told me in the interview that she synthesized well over 1,000 compounds in her most productive year during that time. Well she has only been with us for a couple of months now getting accustomed to our automation lab, and she just synthesized just shy of 2,000 compounds in one day. She did this in a single experiment that she set up and launched on our automated compound synthesis setup that you see in the video below.

Note that this video is a bit old, so the newer lab hardware is more streamlined than this setup. Veteran lab automation nerds might recognize that as the Hamilton Star, fitted with tube twister channels and integrated as a custom Verso picker for 20mL glass scintillation vials. This was a novel technique developed to push the limits of the product, and since a more robust off-the shelf solution has been implemented by the vendor. This is a typical outcome from custom projects like this, the the entire tends to benefit a lot from the work that early adopters do.

Our strategy to automating synthesis is to eliminate the traditional chemistry processes that don't add value. It's undeniably valuable to make and test as many compounds as possible because even bad compounds provide useful data to paint a larger picture. The part of the traditional pharma R&D process that doesn't have value is largely in processing newly synthesized compounds for long term storage and screening, processes typically referred to as compound management. We use a Just In Time inspired strategy to synthesize new compounds on the fly by storing the building blocks. In doing this, we don't need to purify and store finished compounds, because we can always make more as long we have the building blocks.

We store our building blocks in solution, and many more traditional chemists have told us that can't be done. They are right in some cases, but that's why we still keep them around. We focus on the synthesis reactions and reaction sets that we successfully can fit into our synthesis platform. It's not perfect, but we still have what feels like limitless options. I'll let you know when we run out of work.

so rather than purifying and storing finished compounds before testing, synthesis and analyze on the fly. We screen without purifying. Risky? perhaps. Some chemists would think we are mad, and that we can't accurately test functionality that way. Again, they right, it's not 100% perfect. But that hasn't stopped us from finding new hits on our system. All we have to do is supply a worklist holding the appropriate instructions through the software interface, and we can begin synthesizing new compounds. Very few of the vast number of compound tested will one day turn into a new drug.

We do screening assay and data acquisition in-line by integration of LC-MS analysis in the automation setup you see in the video above. This has allowed us to create hands-off workflows that include compound synthesis, QC of the reactions, and functional screening at the click of a button. The reason we like the mass spectrometer as a means of analyzing our experiments is because it provides a detailed view of the chemistry that happened in that experiment. We can tell if our desired compound was synthesized, and we can also tell if it bound to an appropriate ligand. More importantly, we can sort through the sea of returned data to sift through impurities, meaning that purification of the compounds after our synthesis reaction is no longer required.

Now as you can imagine, those traditional medicinal chemists will often scoff at all of this. Partly because the tech still isn't perfect. However a good scientist will always find the flaws, at times to a fault. Perhaps some scientists are scared that they will be replaced by lab automation, but this largely is incorrect for the time being. The automation doesn't design chemistry experiments, it prepares physical research assets according to instructions the chemist provides. However a common bottle neck in this process typically starts with the software interfaces that labs use to house and launch experiments.

Using an approach to chemical synthesis like this, the challenge becomes your reaction enumeration. For non-chemists (like me), this means the challenge is to have a software process that can assess and design chemical reactions as efficiently as a human. Chemists know how the molecules go together, but how can we train software systems using rules and machine learning to help? This is certainly a challenging and fascinating topic, but that's a post for another time.