[cf. Scannell & Bosley2016] Target-based drug discovery is the dominant paradigm of drug discovery; however, a comprehensive evaluation of its real-world efficiency is lacking.

Here, a manual systematic review of about 32,000 articles and patents dating back to 150 years ago demonstrates its apparent inefficiency. Analyzing the origins of all approved drugs reveals that, despite several decades of dominance:

only 9.4% of small-molecule drugs have been discovered through “target-based” assays. Moreover, the therapeutic effects of even this minimal share cannot be solely attributed and reduced to their purported targets, as they depend on numerous off-target mechanisms unconsciously incorporated by phenotypic observations.

The data suggest that reductionist target-based drug discovery may be a cause of the productivity crisis in drug discovery. An evidence-based approach to enhance efficiency seems to be prioritizing, in selecting and optimizing molecules, higher-level phenotypic observations that are closer to the sought-after therapeutic effects using tools like artificial intelligence and machine learning.

…Swinney & Anthony2011 showed that, despite the disproportionate dominance of reductionist target-based drug discovery, most of the first-in-class drugs approved 1999^–₉2008_16ya were discovered by phenotypic approaches. This could have inaugurated a rejuvenation of phenotypic drug discovery; alas, it is still being sidelined by a focus on target-based drug discovery^{144, 145} and is viewed merely as a complementary approach for discovering novel mechanisms of action and first-in-class drugs.^{54, 146}

…While the aforementioned analysis conducted by Swinney & Anthony2011 was seminal in revitalizing phenotypic drug discovery, the period they analyzed (1999^–₉2008_16ya) was so limited that their conclusions, apart from being restricted to the suggestion that phenotypic drug discovery may be more efficient for discovering first-in-class drugs, were subsequently challenged by an analysis that had assessed a longer time frame (Eder et al 2014). Moreover, Swinney & Anthony2011’s analysis did not cover the golden period of traditional drug discovery at all; thus, it is not suitable for comparing the real-world contributions of traditional and target-based drug discovery. Consequently, I expanded the analysis of to include all approved drugs and tried to increase the accuracy and objectivity of the analysis.

…2.1.2. Method: …An example of the application of these definitions can be helpful in order to better gauge the objectivity of the conclusions. Captopril is known by many as one of the early drugs discovered “rationally.” To objectively assign its discovery origin to phenotype-based or target-based, the definitions of these terms given above were followed exactly. In Supplement 1, these quotes from the discoverers of captopril themselves have been cited: “In 1968, Dr. Y. S. Bakhle demonstrated that dog lung ACE was inhibited by a mixture of peptides from the venom of the Brazilian viper Bothrops jararaca…This exercise was not completely in vain; it showed how rare, indeed, were specific inhibitors of ACE, it also demonstrated that these, whether designed or stumbled upon, could be readily identified using a simple guinea pig ileum test system developed by Dr. Rubin and his colleagues. Success in this simple in vitro test was also highly predictive of activity in vivo, including antihypertensive activity…The key result with this prototype compound, however, came in Dr. Rubin’s guinea pig ileum test. Unlike the 2,000 or so random compounds that we had previously tested, succinyl-l-proline had the properties of a specific ACE inhibitor: it inhibited contractile actions of angiotensin I and potentiated those of bradykinin, without having any effects on contractile actions of angiotensin II or those of several other smooth muscle agonists.” (Cushman & Ondetti1991, emphases added) As is evident in the quote, although the discovery took place with a “target” protein in mind, the selections and “the first observation that has related a therapeutic class to the therapeutic effect” were completely based on “observing the effects of molecules on phenotypes.”

…In investigating the discovery origins, I highly prioritized the accounts of the initial reporting discovery papers (denoted as “From the discovery paper”:) and afterward, other narrations from the discoverer(s) themselves (denoted as “From the discoverer(s)”).

…2.1.3. Results and Discussion: Out of the 1,310 US FDA-approved drugs, 69 were endogenous-based biopharmaceuticals and 97 were other biologic drugs. Out of the 1,144 remaining small-molecule drugs, 123 (10.75%) were discovered by target-based drug discovery and 1,021 (89.25%) by phenotype-based approaches. Despite the dominance of small-molecule target-based drug discovery in the last 40 years, it represents a meager share of the currently used drugs (123 drugs (9.39%)), in contrast to phenotype-based approaches (1,021 (77.94%) vs) (Figure 2a). This disproportionate disparity holds up even when only the drugs approved after 1995 are taken into account (123 (17.30%) vs 438 (61.60%)) (Figure 2b). Although the share of target-based drug discovery from the drugs approved each year seems to have grown over time, there has not been a single year in which it was not surpassed by the share of phenotypic drug discovery (Figure 3).

Reviewing the discovery origins provided in Supplement 1 demonstrates that, contrary to the textbook procedure of drug discovery which has been the standard of the past few decades, a substantial portion of the most important drugs used in the clinic have originated from approaches that many drug hunters nowadays would consider “otherworldly” and “taken place on another planet”.

Most of the approved drugs have originated from emphasizing and using highly predictive phenotypic models, like predictive animal models, ex vivo systems, or cultures of bacteria, fungi, and protozoa (about 700 drugs including most of the drugs in these categories: observing nonhuman or ex vivo phenotypes, mechanism of action-informed phenotypic observations, and observing phenotypic effects of endogenous molecules). This is evident in the discovery origins of many drugs including isoproterenol, propranolol, cimetidine, ethosuximide, purine analogs, cyclosporine, pentamidine, omeprazole, propofol, paclitaxel, topotecan, ivermectin, topiramate, leflunomide, sirolimus, ezetimibe, cinacalcet, and tecovirimat.

Many antimicrobial agents that were discovered by observing infected animals or the cultures of the microbes themselves, even during screening uncharacterized mixtures like soil samples, are other notable examples in this regard, like tetracyclines, penicillins, cephalosporins, pyrimethamine, chloroquine, vancomycin, rifampin, clavulanic acid, pretomanid, retapamulin and lefamulin, spinosad, and bedaquiline.

This attention to the predictivity of models was so high that in many cases, drug hunters used themselves as models and self-experimented with molecules, exemplified by lidocaine, amphetamine, cromolyn, bisacodyl, ingenol mebutate, and bremelanotide.

The drugs discovered based on serendipity and sagaciously observing nonhuman or ex vivo phenotypes can also be considered to be indebted to this attention toward predictive models, including bacitracin, ticlopidine, valproic acid, warfarin, meprobamate, vinca alkaloids, dipyridamole, diazoxide, cisplatin, etomidate, naftifine, glatiramer acetate, imiquimod, thiazolidinediones, levetiracetam and piracetam, and dapagliflozin and empagliflozin. Such serendipitous discoveries would be impossible in target-based drug discovery’s biochemical affinity assays where the output is merely the affinity of a molecule for a specific protein.

A minor portion of the discovery origins (the 10.6% human phenotypes plus the 6.8% historically used) can be traced back to serendipity and sagaciously observing the therapeutic effects of substances in humans. Examples include chlorpromazine (the prototype of almost all antipsychotics), imipramine and iproniazide (the prototypes of almost all antidepressants), acetaminophen, corticosteroids, disulfiram, several diuretic and diabetes medications, methyldopa, minoxidil, flibanserin, gemfibrozil, nabilone, anagrelide, sildenafil, azelaic acid, hydroquinone, and memantine. It is fair to add to these cases the many drugs discovered based on historical observations, including metformin, digoxin, artemether, podofilox, ingenol mebutate, fingolimod, eribulin, and spinosad. These cases highlight the importance of sagacity in clinical settings.

…This insight can be inferred from the gleaned data that the “recent increase in productivity”^47,48 can be traced back more to the adaptation of the pharmaceutical industry to its failures rather than addressing their fundamental causes. For example, efforts have been redirected from the unaddressed challenge of complex CNS disorders toward the more reducible rare and monogenic disorders^7,48,49,257 or deriving analogs of drugs discovered decades ago, for example, sarecycline, eravacycline, omadacycline plazomicin, remimazolam, and lumateperone (Supplement 1). Even in some cases, drugs that were not brought forward to the market decades ago have been taken off the shelf and contribute to the apparent “increase in productivity.” For example, rifamycin was originally discovered in 1963 using a phenotypic screen but was not developed further in the US; rather, it was instead optimized to rifampin which the US FDA approved in 1971. Anyhow, in 2018, rifamycin itself was approved by the US FDA, and the drug was introduced to the US market. Other similar examples include triclabendazole, artesunate, moxidectin, and tafenoquine. The peculiarity of these cases of returning to decades-old structures and molecules can be illuminated by heeding that the drug-like chemical space is estimated to be so vast, up to 10⁶⁰ molecules, that novel structures ought not to be scarce.

It is also fair to add to the disproportionate contribution of phenotypic observations the numerous post-approval labeled and off-label indications that have been added for many approved drugs. In many cases, such repurposings and added indications have been based on phenotypic observations made by astute clinicians.^265,266

The systematic review (which included a manual review of 31,027 unique articles) confirmed the hypothesis. Many “target-based” drugs have numerous “off-target” therapeutic mechanisms (Table 2 and Supplement 3). For example, donepezil, which was discovered and developed as an acetylcholinesterase inhibitor, was found to have 40 therapeutic mechanisms independent of acetylcholinesterase…

[Semaglutide’s appetite effect was discovered accidentally in human patients, and the off-target effects on everything from kidney disease to alcoholism were even more unexpected.]

This implies that the contribution of target-based drug discovery to approved drugs is even far less than 9.4%. If it was solely up to the scheme of reductionist target-based drug discovery, none of these “off-target” therapeutic mechanisms would have existed. They have been unconsciously and blindly selected because of the terminal assessments of therapeutic effects and do not necessarily accompany all molecules selected based on their binding to single “targets.” Notably, these counts are inevitably restricted to mechanisms that have been unraveled thus far; it is reasonable to expect that the actual “numbers” would be much higher.³⁵⁴

Derek Lowe commentary: … author, Arash Sadri, has undertaken an extensive review of the origins of all approved small-molecule drugs (1,144 of them). His claim is that only 123 of them were discovered by purely target-based assay methods, and that the rest have to be described as discovered through phenotypic means. He notes that the share of target-based drugs in new approvals has grown over the years, but that there has never been a year when they surpassed the phenotypic ones.

…Whenever possible, he has relied on the original accounts of people who worked on each particular project. And when you go back in that sort of detail, you find that yes indeed, it looks like an awful lot of drugs have been discovered by means that do not match up with the ways that we’d like to imagine that we use. Indeed, he says that the apparent recent increases in productivity have been more due to “the adaptation of the pharmaceutical industry to its failures rather than addressing their fundamental causes”. The popularity of rare-drug monogenic disease drugs falls into this category, for example.

And even for the drugs whose discoveries were unambiguously target-based, in many cases their eventual clinical utility seems to have been enhanced by unanticipated effects that were observed phenotypically in humans. As Sadri has it, there is a survivorship bias at work, since so many drugs fail their Phase II trials. The ones that get through are not simply the ones that had the most selectivity and the most potency for their stated targets. Many, many drugs are doing something else along with (or other than) their ostensible mechanisms. Potency and selectivity for the stated target are simply not sufficient (by themselves) to make a drug, as anyone with any experience knows, but if we really believed in the “strong form” of the target-based drug discovery paradigm, wouldn’t they be (outside of toxicity failures, of course).

…This is not to say that target-based drug discovery has been of no value—for one thing, many of the tools that have been developed for it are useful in answering other questions as well, phenotypic ones included. But we don’t seem to be discovering as many drugs while using it as we think we are, mainly because we don’t seem to be using it as often as we think we are in the first place. Perhaps an acknowledgment of this would be helpful?