This compound looks too much like a detergent

Great post by Ash Jogalekar. Drug discovery is still a prime example of "competence without comprehension" (link): we build things that work without always understanding how or why.

When a mechanism of action (MOA) is known and a target "validated", exploring compounds in the vicinity of known compounds that address that target with that MOA is the only place where we can pretend to know what is going on and hope we can build locally continuous predictive (qSAR) models. For any other situation, we are navigating a super-high-dimensional, rugged landscape in the dark without night vision.

When you are exploring this unknown environment and have little data, you come up with rules of thumb based on what has worked so far. Lipinski's rule of 5 is exactly that: a rule of thumb. Remarkably, Lipinski was able to extract some common features of successful small molecules 28 years ago from his medicinal chemistry experience, not using sophisticated machine learning or AI algorithms. Obviously the field has moved way beyond this rule of thumb, escaping flatland (link). That's why a compound that looks like a detergent may actually be a good candidate.

In the end, all compounds destined for greatness have to face the reality of biology, in animals and humans. That's where we are still throwing "sh*t" at the wall to see what sticks. As data, models and AI progress, we may be able to do away with rules of thumb and be good enough at (explainable) prediction that the amount of "sh*t" we need to invest in and throw at the wall to find one piece that sticks is reduced by orders of magnitude. Ideally, most of what is thrown at the wall sticks. Ash Jogalekar thinks it is not going to happen. We are certainly not there yet, but the drastic reduction in peptide and protein syntheses required to find a hit shows that progress is coming in places that were not on the radar 5 years ago.

"Drug discovery scientists are usually frustrated when they cannot find that one rule or metric to rule them all, but that makes drug discovery a discipline without end, a discipline where novel discoveries are always possible and where the opportunities are endless. What more could a scientist want?"

Mark Murcko, Derek Lowe, Jim Wikel, Aaron Schacht