Technical Debt in Drug Discovery

Almost every resource on Agile practices such as scrum and kanban is aimed at software developers. There are good reasons for that, but there’s no reason why those practices can’t be used for other types of knowledge work like drug discovery. Knowledge work is knowledge work, regardless of the substrate. It’s an iterative cycle of design-make-test-analyze. Software people write code at a computer and tinker with it until it’s a usable app; People like me make molecules in a lab and tinker with them until it’s a usable drug.

If you read through a book on managing software development projects, it’s remarkable what you can learn if you just swap of the names of individual roles (programmer, tester, project manager, etc.) for other names (associate scientist, principal investigator, group leader, etc.). Aside from unfamiliar titles and roles, there are other unfamiliar terms. Some are specific to the software industry, but some are more general. One term in particular describes the shortcuts and uncorrected errors that come back to haunt you later. That’s called technical debt. For software people, that’s usually sloppy, undocumented, non-standardized code. It's hard to follow. It’s bloated, irregular, and buggy. Sure, it’s faster to write bad code, and if it works well enough you can deploy it to your customers. But it’s risky because it introduces unnecessary fragility that can cause enormous headaches later on should something go wrong.

Since the process of knowledge work is very similar across many industries, I started to think about what technical debt means to medicinal chemists like me. Just like in software development, medchem technical debt is invisible and silent, a booby trap of our own making waiting to be sprung. And just like software development, accruing technical can speed along the initial stages of a project, but months or years later the awful truth inevitably surfaces. The necessary re-work could be minor, but perhaps the entire project is put in jeopardy. There could even be billions of dollars at stake!

Below is my partial list of medchem technical debt items sorted into broad categories. Note that these apply to traditional small molecule projects and not necessarily new modalities such as stapled peptides, millamolecular compounds, or antibody-drug conjugates. Note also that many of these items are interrelated. For example, over-reliance on a small subset of synthetic tools can lead to long, inelegant, unscaleable synthetic routes. Another example is how the over-optimization of selected properties can lead to flagrant violations of good drug design principles, especially molecular weight, hydrogen bond donor/acceptor count, and solubility.

Planning

• Inadequate literature searches lead to missed opportunities to find easier synthetic routes and head off potential intellectual property issues
• Long, inelegant, unscaleable synthetic routes are often the result of the above point, and create further problems
• Compounds with known structural liabilities such as PAINS, PrATs, and other problematic functional groups
• Chemotypes known to be problematic within your company or research group
• Flagrant violations of good drug design principles such as cLogP, molecular weight, polar surface area, hydrogen bond donor/acceptor count
• Over-reliance on a small subset of synthetic reactions is addressed in many journal articles. For an example, see this reference.

Synthesis

• Poor laboratory notebook keeping practices are potentially the most costly in terms of lost time and frustration. In the worst case, it could cost billions in sales if when a blockbuster drug's patents are challenged in court.

• Poor lab technique gives a false negative result on synthetic viability
• Purification issues leave open the possibility of impurities that skew assay results

Testing & Analysis

• Unrecognized assay drift will lead you to believe things about your compounds that just aren’t true
• Ignoring the error limits of assays gives the false impression that one compound is better or worse than another when in reality they’re not distinguishable from one another
• Over-optimization of selected properties (e.g., potency and selectivity) at the expense of everything else is a parochial viewpoint that ignores the complex interconnectedness of the drug discovery process
• Check out Garrett Hardin’s Filters Against Folly and Henry Hazlitt’s Economics in One Lesson for the same message in other areas of research

There are many more that aren’t in this list. You can probably come up with a bunch yourself. If you’re not a medicinal chemist yourself, try thinking of examples of technical debt in your particular area of knowledge work. If you are a medicinal chemist, you can go deeper down the rabbit hole by reading this Drug Discovery Today article from scientists at AstraZeneca.

Of course there’s always a balance to be struck between speed and quality. Research is complex, messy, and uncertain. Although some technical debt items are non-negotiable, others have some ‘wiggle room’ as long as you’re aware of the possible consequences and have a plan to fix it later (assuming the problems are fixable to begin with). Edmond Lau makes the same point in his new book The Effective Engineer, another guide on software development that transfers easily to drug discovery.

So there you have it. If you’re taking some time off over the holidays, why not read through a software development book or two? I think you’ll find many striking parallels that transfer easily to your own work.