For most of the 2010s, PROTACs were an elegant idea with an ugly pharmacology problem. The concept — hijacking the cell's own protein disposal machinery to destroy disease-causing proteins — was scientifically compelling. But PROTAC molecules are large (typically 700–1,100 Da), violate most traditional drug-likeness rules, and behave unpredictably in the body. Designing one that works in a test tube is difficult. Designing one that works in a patient proved nearly impossible with conventional medicinal chemistry.
That is changing, and AI is the reason.
The Mechanism
PROTACs (Proteolysis Targeting Chimeras) are bifunctional molecules. One end binds a disease-causing protein. The other end binds an E3 ubiquitin ligase — one of the cell's protein recycling enzymes. When both ends are engaged, the PROTAC brings the target protein into close proximity with the ligase, which tags it with ubiquitin, marking it for destruction by the proteasome.
The advantage over traditional inhibitors is fundamental. A conventional drug blocks a protein's function by sitting in its active site — but only while the drug is present and at sufficient concentration. A PROTAC destroys the protein entirely, and because the PROTAC itself is recycled after each degradation event, a single molecule can catalytically eliminate many copies of its target. This catalytic mechanism means PROTACs can work at lower doses, overcome resistance mutations, and target proteins that have no conventional binding pocket — the so-called "undruggable" proteome.
The disadvantage is molecular complexity. A PROTAC has three components: the target-binding warhead, the E3 ligase-binding ligand, and a chemical linker connecting them. The linker's length, flexibility, and chemistry critically determine whether the ternary complex (target-PROTAC-ligase) forms productively. Small changes in linker geometry can mean the difference between potent degradation and no activity at all.
Clinical Momentum
As of early 2026, at least 30 PROTACs have entered human clinical trials, with the vast majority in Phase I or II. The field's most advanced candidate is vepdegestrant (ARV-471), an estrogen receptor degrader co-developed by Arvinas and Pfizer for hormone receptor-positive breast cancer. The FDA accepted Arvinas's New Drug Application in August 2025, with approval expected by June 2026. If approved, vepdegestrant would be the first PROTAC to reach the market — a watershed moment for the entire targeted protein degradation field.
Bristol Myers Squibb's BMS-986365 (CC-94676), an androgen receptor PROTAC for prostate cancer, became the second degrader to enter Phase III trials in 2025. The ARV-766 program, also targeting the androgen receptor, reported encouraging Phase I/II data in metastatic castration-resistant prostate cancer.
Beyond oncology, the pipeline is expanding. Kymera Therapeutics has PROTACs in clinical development for immunology and inflammation, targeting IRAK4 and STAT3. Nurix Therapeutics is developing degraders for Bruton's tyrosine kinase (BTK) in B-cell malignancies, with early data showing activity against BTK mutations that confer resistance to conventional inhibitors — precisely the kind of scenario where PROTACs' mechanism of action provides an advantage.
The AI Inflection
The PROTAC design problem is almost perfectly suited to AI. The combinatorial space is enormous: for any given target, there are hundreds of possible warheads, dozens of E3 ligase ligands, and thousands of linker variations. Testing each combination experimentally is prohibitively slow. But the structure-activity relationships are learnable — there are patterns in which linker geometries produce productive ternary complexes, and AI models can identify those patterns from existing data.
Several approaches have converged. Generative models trained on existing PROTAC structures can propose novel linker chemistries that are likely to produce active degraders. Molecular dynamics simulations, accelerated by machine learning, can predict whether a proposed ternary complex is stable. And structure prediction models like Boltz-2 can now model the three-body interaction between target, PROTAC, and ligase with reasonable accuracy.
A 2025 review in Pharmaceuticals catalogued the state of AI-assisted PROTAC design, documenting dozens of studies where machine learning models successfully predicted degradation activity, optimised linker geometry, or identified novel E3 ligase recruiters. The most impressive results came from generative models that proposed PROTACs with activity against targets where no prior degrader existed — true de novo design rather than optimisation of known scaffolds.
Molecular Glues: The Parallel Track
Alongside PROTACs, molecular glues have emerged as a complementary approach to induced proximity. Molecular glues are smaller molecules (typically 300–500 Da) that stabilise interactions between a target protein and an E3 ligase without the need for a bifunctional linker. They offer better pharmacokinetic properties — oral bioavailability, blood-brain barrier penetration — but are harder to design rationally because their mechanism is less modular.
The paradigm case is thalidomide and its derivatives (lenalidomide, pomalidomide), which were discovered to work as molecular glues for the E3 ligase cereblon decades after their initial clinical use. The challenge is moving from serendipitous discovery to rational design. AI is making progress: computational screens can now identify potential molecular glue interactions by modelling protein-protein interfaces and predicting where small molecules might stabilise transient contacts.
Partnerships worth over $200 million were signed in 2025 alone across the targeted protein degradation space, with the majority of large pharma companies now invested in molecular glues, PROTACs, or both.
Beyond the Proteasome
The next frontier is expanding the degradation toolkit beyond the ubiquitin-proteasome system. Autophagy-targeting chimeras (AUTACs) and lysosome-targeting chimeras (LYTACs) use different cellular recycling pathways to degrade targets that are inaccessible to PROTACs — including extracellular proteins and protein aggregates. These approaches are earlier-stage but conceptually powerful: if you can co-opt any of the cell's multiple degradation pathways, the fraction of the proteome that is truly "undruggable" shrinks dramatically.
A 2026 publication in Biomaterials described nano-PROTACs — PROTAC molecules delivered via engineered liposomes that can target specific tissues and cell types. The study demonstrated targeted degradation of hexokinase 2 in tumour cells, enhancing chemotherapy sensitivity. This kind of tissue-targeted degradation could address one of the persistent concerns about PROTACs: that systemic protein degradation might produce toxicities in healthy tissue.
The Road to a $10 Billion Market
If vepdegestrant receives FDA approval in mid-2026, it will validate not just a single drug but an entire therapeutic modality. The addressable market for targeted protein degradation — encompassing PROTACs, molecular glues, and next-generation chimeras — is estimated to exceed $10 billion by 2030, driven by applications in oncology, immunology, neurodegeneration, and infectious disease.
The AI component is not optional. The molecular complexity of degrader design means that traditional medicinal chemistry, even at its best, can explore only a fraction of the relevant chemical space. AI-driven design expands that exploration by orders of magnitude. The companies that will lead the next wave of degrader therapeutics are those combining deep biology (understanding of E3 ligases, ternary complex dynamics, tissue-specific degradation) with sophisticated computational platforms.
The PROTAC concept is 25 years old. It took two decades of chemistry to produce the first clinical candidates. AI may compress the next generation of degraders into a fraction of that time.