Eighty percent of the human proteome is classified as undruggable. The surfaces are smooth. The binding pockets are missing. Conventional small molecules slide off them like water off wax. In the first half of 2026 alone, more than €8 million in seed and early-stage capital flowed into a single premise. The proteins traditional drugs cannot touch are precisely the ones a new class of molecules, designed by artificial intelligence, was built to reach.
AI-driven platforms from startups like Ternary Therapeutics and Amphista are shifting molecular glue discovery from serendipity to systematic engineering, but the field has already buried two BRD9 degrader programmes from Foghorn and C4 Therapeutics.
For investors, the question is not whether molecular glues can drug the undruggable. The science is sound. The question is which platform companies will survive the transition from preclinical promise to clinical data.
Targeted protein degradation market
Driven by molecular glues and PROTACs expanding beyond oncology into autoimmune and CNS indications. · MarketsandMarkets, 2026
What makes a protein undruggable
The lock-and-key model that gave pharmacology its most productive metaphor works well for enzymes and receptors, proteins with deep pockets where a small molecule can dock and change behaviour. But roughly 80% of the proteome consists of transcription factors, scaffolding proteins, and intrinsically disordered regions. They have no pockets. They fold only when they meet another protein. A conventional inhibitor has nothing to latch onto.
Molecular glues solve this differently. Instead of blocking a protein, they act as a coupler, binding simultaneously to a target protein and an E3 ubiquitin ligase, the cell's natural tagging machinery. The ligase then marks the target for destruction. The protein is not inhibited. It is removed.
Three drugs already work this way: thalidomide, lenalidomide, and pomalidomide, collectively known as IMiDs. They recruit the CRBN E3 ligase to degrade transcription factors in multiple myeloma and reached peak combined sales of $16 billion. The mechanism is validated. The challenge is extending it beyond the handful of ligases and targets these molecules happen to hit.
Position: AI turns serendipity into engineering
Every molecular glue on the market today was found by accident. Thalidomide's mechanism was not understood until decades after its approval. Design was retrospective, not prospective.
Ternary Therapeutics, a London-based startup incorporated in October 2024, is building the platform to change that. The company raised €4.1 million (£3.6 million) in a seed round led by daphni, with participation from Pace Ventures, the i&i Biotech Fund, and the UK Innovation & Science Seed Fund. Its pitch is straightforward: combine physics-informed AI with rapid experimental feedback to design molecular glues deliberately, rather than screening for them.
Ternary's platform runs molecular dynamics simulations to predict how proteins move and interact at atomic resolution, then proposes small molecules that could induce a stable ternary complex, bringing together the glue, the target, and the E3 ligase. Those candidates are tested in the lab within days, and the results refine the next prediction cycle. The company already has a preclinical pipeline in inflammatory and neuroinflammatory diseases and has signed multiple research collaborations with pharma partners.
"For decades, molecular glues have been one of the most powerful ideas in drug discovery, but also one of the least predictable," said Dr Chris Tame, co-founder and CEO of Ternary. "We have shown that by combining physics-informed AI with tight experimental feedback, we can design these molecules intentionally rather than waiting to discover them by luck."
The approach attracted the attention of Ian Taylor, former president of R&D at Arvinas, who joined Ternary's board as part of the round. Arvinas is one of the best-known companies in targeted protein degradation.
Amphista Therapeutics, a Cambridge-based biotech, is further advanced clinically. Its lead candidate AMX-883, an orally bioavailable molecular glue degrader of BRD9, is on track to enter a Phase I trial for acute myeloid leukaemia (AML) in the second half of 2026. The compound targets BRD9 through a completely novel E3 ligase, DCAF16, distinct from the CRBN and VHL pathways that dominate the field. Preclinical data presented at AACR 2026 showed that AMX-883 induces near-complete degradation of BRD9 within two hours and demonstrates synergistic efficacy with venetoclax, the standard-of-care BCL-2 inhibitor, while preventing the emergence of resistance.
Its Eclipsys platform, which integrates cryo-electron microscopy, geometric deep learning, and cheminformatics, has already produced degraders against additional novel ligases including DCAF11, targeting KRAS G12D, one of the most sought-after targets in oncology.
Counter: clinical gravity applies to everyone
The TPD field has generated extraordinary preclinical data. Translating that into clinical benefit has proven harder.
Two BRD9 degrader programmes preceded Amphista's. Foghorn Therapeutics advanced a CRBN-based BRD9 degrader into a Phase I trial for synovial sarcoma in 2021. The FDA placed a partial clinical hold in April 2023, and Foghorn discontinued the programme. C4 Therapeutics followed with its own BRD9 degrader in 2022, calling time on it later in 2023 after high levels of target degradation failed to translate into sufficient efficacy.
Both were PROTACs rather than molecular glues, and both relied on the CRBN ligase. Amphista explicitly designed around that limitation by recruiting DCAF16 instead. But the precedent matters: degrading a protein in a dish and degrading it in a patient at a therapeutic index are two different problems, and the second one has a higher failure rate than the first.
What else could go wrong
− Platform valuation risk. AI-driven drug discovery companies command premium valuations before clinical data. If Ternary's platform delivers candidates that fail in early trials, the valuation multiple on "platform story" compresses and the company may not have the cash runway to pivot. − Undruggable means hard to drug for a reason. Transcription factors and scaffolding proteins have evolved surfaces that resist small-molecule binding. Making a glue that sticks to them while avoiding off-target degradation of unrelated proteins is a selectivity problem the field has not yet solved at scale. − Manufacturing complexity. Ternary complexes are more structurally intricate than conventional drug-target interactions. Scale-up and formulation for oral bioavailability, Amphista's claim for AMX-883, adds an engineering variable that has killed otherwise promising modalities before.
Why this matters for the private markets
The TPD space illustrates a pattern that recurs across deep-tech investing: a validated mechanism creates a race to build the platform that can exploit it systematically. The winners are not necessarily the first to publish; thalidomide was discovered in the 1950s. The winners are the ones that combine computational design capability with experimental throughput and pragmatic target selection.
Three signals worth tracking:
Ternary Therapeutics names a lead candidate and declares an indication for clinical development. The platform thesis becomes falsifiable.
Its AMX-883 Phase I data in H2 2026. If safety and early efficacy hold, it validates the DCAF16 ligase strategy for the entire field.
Big pharma deal activity. BMS and Merck KGaA have already placed bets on Amphista. A follow-on licensing deal for a different ligase or a new platform entrant would signal institutional conviction.
PROTAC vs molecular glue divergence. If vepdegestrant (Arvinas) wins FDA approval in 2026, capital will flow disproportionately toward the degrader modality that can point to a commercial product.
The total addressable opportunity is large enough that multiple platforms can coexist through Phase II. But the capital required to run controlled trials at that stage, typically $50-100 million per indication, means most of the current seed and Series A companies will not make the transition without either strong Phase I/II data or a pharma partnership. The discipline for an investor in this space is distinguishing platform companies that are building a repeatable discovery engine from those that are essentially a single-target bet dressed in AI language.
