Which targets are overheated?

In some cases it can be rational to go after a crowded target if you think you have a differentiated approach to improve the modulation of known, validated biology (e.g., longer dosing enabled by PCSK9 siRNAs), or the market is simply so large that it can support many entrants (e.g., GLP-1 receptor agonists in obesity). Yet in some cases we see companies piling into novel targets before they’ve been validated by clinical efficacy or market demand.

Prior analyses have focused on simple counts of active drug programs by target as an indicator of crowdedness. We were interested in adding another layer: How rapidly do targets accumulate new clinical programs? And how does the rate of accumulation of new drugs against a target track with what we'd expect relative to the level of validation?

A typical target attracts new clinical-stage molecules at a steady cadence of about 0.3 per year after its first clinical entry, with a modest increase in annual starts over time as development activity matures. However, for some targets we see new program starts lag clinical and regulatory validation milestones, and for other targets new clinical starts race ahead of clinical validation. For instance, PCSK9 had significant genetic support and development activity was intense even before clinical and regulatory validation. Other targets only become hot after substantial evidence has accumulated, or never become hot at all despite solid biological rationale.

Regulatory approval is the most important form of validation, and the US FDA is the most important regulator. We should expect that as drug developers become more confident that a target can support approvable drugs, development activity should increase; we see a modest increase in new clinical entry rates for targets leading up to and after the first approval of a drug against that target. In some cases (e.g., Claudin 18) confidence rises before approval, in others it's not until after the approval that target activity rises substantially.

Heat index

Overall, overheated targets concentrate in oncology and hematology. This is expected, oncology has been fiercely competitive for years. Familiar targets like PD-1, EGFR, HER2, VEGF, and BTK rank highly and remain crowded with later-generation follow-ons and bi- and trispecific approaches, ivonescimab, a PD-1 × VEGF bispecific being among the most notable. CD3's position at the top of list reflects its use as the preferred T-cell surface antigen in multispecific T-cell engagers.

Topoisomerase I is an older chemotherapy target, but its crowded pipeline reflects the ADC surge and convergence on Topo1-inhibitor payloads (e.g., deruxtecan) following Enhertu's clinical success.

Not all this activity follows an established success. MSLN and CD47 remain popular despite a difficult clinical record. The mesothelin-directed ADC anetumab ravtansine did not improve progression-free survival over vinorelbine in mesothelioma, and Gilead stopped magrolimab development in hematologic cancers after multiple failures. Yet MSLN and CD47 have 40 and 35 active drugs, respectively, and some developers remain optimistic that a different construct, modality or treatment context can overcome earlier limitations.

CD19 and BCMA’s high rankings are a continuation of the long trend of cancer targets gradually moving over to immunology, and in particular the new wave of CAR-T-based cell-depletion strategies (kicked off by promising early results). About 68% of the CD19 drugs are being developed for immunology indiations, and the majority of these (57%) are CAR-Ts. For BCMA, 48% of the drugs are being developed in immunology. These overlapping counts support an autoimmunity-led explanation for much of CD19’s heat: companies are trying to turn B-cell depletion into a durable immune reset.

Beyond oncology, GLP-1R remains hot as obesity development expands into adjacent indications and combination approaches. JAK1 and S1P1 also rank highly, with activity spanning dermatology, immunology and neurology.

Pipeline trends

We're also interested in seeing how target activity has changed over time to put the current index into context. The largest heat-score increases since 2025 show where competition is moving fastest relative to the evidence.

Next-generation obesity targets like INHBE and ALK7 are fast risers; for instance, Arrowhead has RNAi programs directed against both INHBE and ALK7 in the clinic. An established market for GLP-1s is pulling development into adjacent biology before those approaches are comparably validated. While GLP-1s are already highly effective and unlikely to be displaced, these novel targets are bets that metabolism and body composition can be sculpted in a manner that is complementary to GLP-1 use.

Oncology accounts for several other large increases in target heat. While CD56 and CD18 rose from a low base, others like ALPP and integrin beta 6 are significant risers within the top 200. ALPP's rise has been driven by new engineered T-cell therapies and a T-cell engager. Integrin beta 6 by a rise in ADCs including sigvotatug vedotin and other Pfizer candidates. Another theme in the past year is tumor microenvironment modulation such as PLT012’s attempt to alter lipid-driven immunosuppression in the tumor microenvironment and MST-0312’s approach to stimulating local antitumor immunity through LTBR activation.

What targets are potentially underhyped?

Negative or low heat scores point to targets where the clinical pipeline is smaller relative to the level of validation estimated by our model. Often, it's clear why companies haven't pursued these targets: small markets, delivery problems, uncertain biology, safety concerns and difficult endpoints are common issues. Yet there are interesting targets in this batch that seem plausibly underhyped. We pick out a few molecules below to show some of these low heat targets that seem interesting.

Conclusion

If we zoom out and look at macro trends, we see a clear tendency toward increased target crowding over time. Novel targets account for a decreasing share of new clinical drugs: 13.5% in 2015, versus 6.4% in 2025. About 6.1% of new drugs entering the clinic are going after targets that haven't been tested clinically before, and ~58% are going after targets with an approved drug.

One interpretation of this trend is that, like the movie industry, new developments are so expensive and risky that it's easier to reboot and remake than to try something genuinely new. Another is that this is mostly an oncology related phenomenon driven by the industry's great push into cancer drugs that started about 20 years ago. While target crowding is a problem in all therapeutic areas, it's mainly a problem in oncology where failure rates and competition are both extremely high. Oncology has had a particularly large expansion in new programs over the past decade: annual clinical entrants increased 2.4-fold in 10 years from 2015 to 2025. Over the same time the first-time-target share fell from 13.1% to 5.8%. We've seen some pullback and rotation into other areas like immunology and neurology in recent years as developers begin to question their return on investment. Perhaps the most pessimistic interpretation is that, in many therapeutic areas, single target biology seems to be mined out. Even targets supported by human genetics have produced major disappointments: Novartis’ pelacarsen lowered Lp(a) but missed its cardiovascular endpoint in Lp(a)HORIZON, and Novo Nordisk’s IL-6 inhibitor ziltivekimab missed its cardiovascular endpoint in ZEUS despite lowering inflammatory markers. Both are particularly striking because of strong genetic support linking Lp(a) and IL-6 receptor signaling to coronary heart disease. While we at Convoke are certain that AI (both LLMs/agentic AI and specialized molecular design models) will increase industry efficiency over a medium-term time horizon, our expectation is that it will increase molecule-design and development capacity faster than it increases the supply of genuinely new, clinically useful target ideas. If that is right, better tools will produce more crowding around the same biology. Faster, cheaper, and more robust human data collection is likely necessary to meaningfully increase the supply of new targets.

Methodology

Targets are usually thought of as being validated or not; it's usually a binary description. However, in the absence of an FDA approval or an unambiguously positive adequately powered clinical trial, how do we say a target is "validated"? Can we say that a target is 20%, or 50%, or 80% validated when we don't know if a drug against that target will eventually be approved? FDA approvals are relatively rare events, and not every target with a drug in the clinic has a historical approval for that target. For our heat index, we wanted a score that captured some measure of confidence in validation for a target, or 'progress toward validation' even without an FDA approval. After some experimentation, we came up with an approach that trained a model to answer this question similar to this one: How much does the pattern of development (incl. number of drugs by stage and failure rates) in this target–indication pair (for example, TNF in Rheumatoid arthritis) look like indication–target pairs with a recorded FDA approval in our drug program database?

We used a random forest (RF), a collection of decision trees that each learn small rules about which evidence patterns tend to accompany FDA approvals. Each tree in the forest estimates the frequency of approval-bearing examples in the group a target–indication pair falls into. Random forest models are nice because they are relatively easy to interpret and are simple to implement and train. Specifically, we hid information on whether or not a drug was approved and trained the RF model to predict whether or not there is currently an approved drug for each target–indication pair.

To increase the sample and look at temporal trends, we reconstruct annual pipelines back to 2000, giving the model earlier and sparser views of the evidence. If we look at the validation over time with an example, e.g., TSLP in asthma, we can see that as evidence accumulated and programs advanced our simple model got more confident in the level of validation.

Finally, to calculate the heat score (shown in the heat index table), we take a clinical-stage program count-weighted average of its RF predicted validation scores across eligible indications for each target. We then fit a linear regression relating that score to the log of the number of distinct active clinical-stage drugs. Converting the fitted value back into a drug count gives us the expected number of drugs at that level of validation.

The heat score measures how far the actual drug count is above or below that expectation, on a scale from −100 to +100. Zero means as many drugs as expected for the level of evidence. High scores indicate more active clinical drugs than expected given the target’s modeled validation.