Although the gap is rapidly closing, most drugs in clinical use were still discovered in the 20th century by phenotypic readouts: direct effects on cells, animals or humans without knowledge of mechanism of action. While much of the pharma industry has pivoted back to this approach, the past 40 years makes the molecular, mechanistic approach to drug discovery nearly unavoidable. So like, yes, it can be more effective to trust your eyes and not be biased by looking at the statistics when evaluating basketball prospects, but how can you ignore them when they are RIGHT THERE under your nose?
Still, the best of both worlds might be appreciating polypharmacology, which is a core concept of GeneCentrix’s technology. Polypharmacology means that drugs and drug candidates have multiple target receptors, not just the one for which they are best known. This means that some of their phenotypic effects may come from one of those other overlooked receptors. This also means that their mechanisms of action are complex, especially for those oldie but goodie drugs that were phenotypically identified.
The flip side of this complexity isis repurposing of those older drugs for new uses by revealing and focusing on the overlooked target. A great example of this is Viagra, which was a heart drug that had a marketable, ahem, side effect. While most repurposing success stories were also phenotype based like VIagra (a new phenotype, not a new target, was found), it remains theoretically possible and promising to dissect out a desireable, target-specific effect of existing drugs and work to amplify it with more selective derivatives.
This approach, of course, would require identification of the group of targets hit by a drug, and that’s the challenge, while the members of this group is clearly related by their binding to the drug, they may not be easily related by any other means. Drug targets are commonly proteins, and protein similarity can be measured by calculating their amino acid sequence similarity or comparing their 3D structures. However, protein sequence/structural similarity does not correlate at all with drug binding, with highly similar proteins not binding to the same drug and vastly different proteins binding the same or similar drug compounds. Similarly, inference by chemical structure similarity between drugs is choppy: even changing a single hydrogen can wipe out binding, while completely different chemical structures can bind the same target pocket as long as the contacts between the drug and the protein are preserved.
Brian Shoichet and his protégé Michael Keiser had a clever, simple idea to improve the approach to this issue (funny how the good ideas are invariably also simple, eh?). They reasoned that drug targets should be grouped according to the drugs that bind them, no matter how different they were to each other. This, of course, is correct in the sense that drug binding is an incontrovertible similarity measure, and also that only the binding pocket need be similar, not the rest of the protein. So, once you have these groups, then it becomes more likely that a drug that binds to one of the groups also binds to other members. If you then layer in the close neighbor range of chemical structure similarity, which is fairly reliable, you end up with something close to the polypharmacologic ensemble of drug classes. They then searched this graph of clusters of drug targets with drugs or drug candidates, and voila, unrecognized biologically significant targets emerged, which they proved experimentally in some cases. They called this the SEA or similarity ensemble approach. Seems so simple, but no one had done it before.
Now, it also seems simple to reason to the next layer, which is that those drug targets vary across the tissues of the body, and that the phenotypes will depend almost exclusively on the nature of the tissue in which the drug is active, right? Now, where have we heard that before . . .