Astellas bets on de novo miniproteins to refresh its early pipeline
Astellas Pharma has struck a partnership with Boston-based AI Proteins that could reach $400 million, placing computational drug discovery at the center of its next wave of early-stage research. The agreement gives Astellas access to AI Proteins’ platform for designing miniproteins from scratch, a strategy built to move beyond the limits of biologics borrowed from natural molecules.
The structure of the deal signals more than another licensing transaction. Astellas will take the lead on development, regulatory processing, and commercialization if any candidates emerge from the collaboration, while AI Proteins receives an undisclosed upfront payment and stands to earn milestone-based value as programs advance. In practical terms, the Japanese company is buying access to a discovery engine that aims to generate fit-for-purpose molecules at the very start of the pipeline, before any candidate ever reaches the clinic.
That approach reflects a broader shift in how large drugmakers are thinking about innovation. Instead of relying only on classic business development, Astellas is trying to widen the pool of possible starting points for new medicines, especially as it faces a future shaped by the patent cliff around Xtandi, its cancer franchise anchor. Xtandi generated $6 billion in sales last year, making the need for durable replacement assets impossible to ignore.
A computational platform built to design molecules from first principles
AI Proteins has built its proposition around de novo protein design. Rather than searching nature for a protein that can be adapted to a therapeutic purpose, the company uses a computational platform to craft miniproteins with the intended binding behavior already built in. Chris Bahl, the company’s chief executive and founder, framed the concept plainly: existing biologics are based on molecules that evolved for some other function, while de novo design reverses that logic by specifying the desired therapeutic action first and then constructing a molecule to match.
In the company’s framing, these designed miniproteins have demonstrated binding activity beyond what traditional nature-derived biologics can achieve. Whether that promise translates into clinically useful medicines will depend on the usual filters of potency, selectivity, manufacturability, safety, and pharmacokinetics. Still, the appeal is obvious. A molecule designed from the outset for a therapeutic target may offer more control over size, affinity, and molecular behavior than a protein repurposed from biology’s existing inventory.
For a company like Astellas, the attraction is strategic as much as scientific. A platform that can generate novel biological matter widens the search space for targets that may be hard to address with conventional small molecules or antibodies. It also supports an early research model in which AI-assisted design compresses the distance between target concept and candidate generation. The practical consequence is a better chance of finding differentiated assets before competitors crowd the same biological territory.
Astellas leans further into AI after years of internal adoption
Astellas has been building toward this type of partnership for several years. The company began incorporating artificial intelligence into development processes in 2019, using a “human-in-the-loop” model to organize data collection and research operations. The AI Proteins collaboration extends that strategy into a more specialized scientific layer, where algorithmic design is not just supporting the workflow but helping create the actual therapeutic starting material.
This is a notable escalation in how pharma is using AI. Many companies first use machine learning to improve trial operations, search literature, or refine chemistry. Astellas is now moving into protein design itself, a domain where computational methods can influence the physical architecture of a candidate molecule before bench work begins. If the platform performs as intended, the gain is not merely speed. It is access to molecules that might never have been conceived through standard discovery routes.
Morten Sogaard, senior vice president of Astellas Innovation Lab, said the collaboration is meant to strengthen the company’s early-stage research pipeline through access to novel biology, differentiated technology, and new therapeutic possibilities. He also tied the partnership directly to the company’s pipeline-led growth strategy. That framing fits the commercial reality. Astellas needs long-duration innovation to offset impending pressure on established revenue, and partnerships like this are designed to create future options rather than immediate sales.
AI Proteins turns a series of partnerships into validation
For AI Proteins, the Astellas deal adds to a growing list of discovery-focused collaborations. The company signed a separate arrangement with Bristol Myers Squibb at the end of 2024, a pact also described as potentially worth $400 million. Earlier that year it entered an R&D collaboration with oral biologic platform company Vivtex, and it has also explored radioligand therapies with the University of Missouri through a collaborative research agreement.
The recent deal flow suggests that large pharma companies are willing to treat the platform as a serious source of candidate generation rather than a speculative technology demo. AI Proteins’ fundraising has followed a similar arc. The company closed a $41.5 million Series A in November 2025 after an initial $18.2 million seed round in 2022. For a young discovery company, the combination of capital and repeat partnering helps validate the premise that de novo protein engineering can support multiple therapeutic programs, not just a single flagship project.
Still, the road from computational design to approved medicine remains long. A platform can impress in binding assays and early screens while still failing on the less glamorous problems that decide drug fate: stability, delivery, immunogenicity, and translational relevance. Miniproteins may be built with precision, but the body applies its own rules once those molecules enter circulation. Any candidate emerging from this collaboration will still have to survive that biological gauntlet.
The real test comes after discovery
Astellas’ move makes sense because it shifts risk toward the earliest phase, where scientific optionality is highest and capital can still be deployed strategically. The company is not just buying a single asset; it is buying a mechanism for generating a stream of possible assets. For a pharmaceutical company managing long-term portfolio renewal, that can be more valuable than chasing late-stage fill-ins with crowded mechanisms.
The deeper question is whether de novo-designed miniproteins can reliably produce medicines that are not only novel but clinically durable. Platform partnerships often succeed at opening the door to discovery and fail at the harder task of making the resulting molecules into therapies physicians can actually use. Astellas will own the downstream burden if any programs progress, including development, regulatory review, and commercialization. The scientific promise is early and real; the clinical proof, as always, will come much later, molecule by molecule, target by target.