AI could help cell and gene therapy manufacturers gain a deeper understanding of the complex production processes used to make their products and predict problems before they occur.
A team led by researchers at Northeastern University College of Science in Boston made the case for AI use in a recent paper, arguing that the variability inherent in cell and gene therapy production can be difficult to manage using conventional tech.
Lead author, Jared Auclair, PhD, dean of the College of Professional Studies at Northeastern, tells GEN, “Unlike monoclonal antibodies or recombinant proteins, cell and gene therapies are living or highly complex biological products, making them inherently more variable and difficult to manufacture consistently.
“Every step, from sourcing starting material to manufacturing, analytical testing, storage, and delivery, can influence the final product,” he adds.
Understanding complex, multi-parameter interactions is exactly the sort of challenge at which AI excels, Auclair says, citing the ability to identify critical process attributes as an example.
“AI has the potential to transform cell and gene therapy manufacturing by moving from reactive to predictive manufacturing. Machine learning can optimize process parameters, predict batch failures before they occur, enable digital twins to simulate manufacturing changes, and strengthen quality control through real-time monitoring and anomaly detection,” he adds.
“At Northeastern, our research at the intersection of the Bioanalytical Training Laboratory (BATL), the Center for Bioinnovation and Regulatory Sciences, and AI is exploring how AI can accelerate the development, manufacturing, and regulation of advanced therapies,” Auclair says.
Not plug-and-play
AI’s potential to spot patterns in data is attractive.
However, biopharmaceutical companies looking to adopt the technology are likely to encounter challenges, according to Auclair, who cautions that setting up an AI-driven manufacturing operation is about more than simply buying the right software.
“The technology is advancing rapidly, but successful implementation depends on having high-quality, well-curated data, digital manufacturing infrastructure, and multidisciplinary expertise spanning biology, engineering, data science, and regulatory science.
“AI is not a plug-and-play solution; organizations must build integrated data ecosystems and governance frameworks that regulators can trust,” Auclair says.
AI adoption is a multidisciplinary challenge and should involve people with expertise in all parts of drug development and production, according to study co-author Rominder Singh, PhD, professor of practice, regulatory sciences, & AI at Northeastern.
“Research conducted through the BATL and the Center for Bioinnovation, led by Professor Auclair, has focused on addressing many of these scientific and manufacturing challenges that are unique to advanced therapies.
“This is precisely why Northeastern’s pioneering work in RegSciAI is so important: it brings together regulatory science and AI to ensure these technologies are both innovative and deployable in real-world biomanufacturing,” Singh says.
In vivo CAR T sounds like the existing class of Chimeric Antigen Receptor (CAR) T-cell therapies used to provide often individualized treatment for cancer. But they’re a new and emerging class of therapeutics with their own challenges and opportunities for manufacturers.
That’s according to Mo Heidaran, PhD, chief scientist at Cellx, who is due to give a talk at the upcoming Bioprocessing Summit in Boston.
“Whether something is a cell or gene therapy, from a regulatory perspective, depends on [the nature of] the product that’s administered to the patient,” Heidaran explains.
“And, in the United States, in vivo CAR Ts are gene therapy products and not cell therapies as some people talk about them.”
The better-known CAR T products are ex vivo, delivered via modification of patient cells, he explains. Whereas this emerging class of therapies involves delivery of a genetically engineered virus or lipid nanoparticle (LNP) that, in some cases, is stably integrated into the patient genome.
According to Heidaran, the risk of integration is higher when viruses are used.
“My colleagues at the FDA want to make sure people understand it’s very important these products must be [designed] to be very specific to the cell type, perhaps based on data about [some of these] therapies having off-target effects,” he says.
Most in vivo CAR T-cell therapies are in very early stages, with none currently approved for patients, although Heidaran says they are increasingly under investigation by larger companies since they are scalable for a wider range of patients. Also, they are believed to be more cost-effective and have similar logistics, as they don’t require lymphodepletion, he adds.
“Essentially the value driver is that you’re pharmaceuticalizing cell and gene therapy since it’s just a vial of the virus or LNP that you can use to treat many patients—almost like a drug or pill,” he says.
Among the challenges for this emerging class is that several ex vivo CAR T-cell therapies are already approved for patients. In vivo CAR T therapies treat some of the same indications, i.e., certain cancers and autoimmune diseases, he says.
“At some point there has to be a decision made by the FDA about how these [new] therapies compare, such as [running] a study or external control as to whether they’re superior or non-inferior to the same or similar approved ex vivo CAR T,” he says.
Other challenges facing this new industry are about batch sizes for manufacturing, as the equipment and processes for treating ten patients are different from needing to treat thousands. Also, he says, in vivo CAR T therapies need to be monitored to look for off-target effects, durability of response, or an immune response by the patient.
“Overall, to develop a safety profile, we need to define what the effective dose is that people are working to, as these therapies may require repeat administration, which is not done with ex vivo-generated CAR T,” he says.
Approximately three million people worldwide struggle with chronic pancreatitis, for which there is no cure. In a study published in Cell Stem Cell titled “Patient-derived organoids reveal ductal dysfunction and CFTR-modulator responses in chronic pancreatitis,” researchers from Salk Institute have developed an organoid platform to uncover the mechanism of chronic pancreatitis development and identify possible therapeutic strategies.
The authors generated 37 organoids from patients who developed chronic pancreatitis spontaneously. The organoids revealed consistent dysfunction in the protein cystic fibrosis transmembrane conductance regulator (CFTR), which was identified as a therapeutic target.
“Though patients can have the same clinical diagnosis of chronic pancreatitis, they can have very different underlying molecular drivers of that disease, which makes treatment especially difficult,” said Dannielle Engle, PhD, assistant professor at Salk and corresponding author of the study. “Our work breaks down a major barrier in the field by establishing an experimental model that preserves patient-specific disease biology and can be used to develop tailored therapies.”
Over the last decade, organoids have become a prevalent tool to bridge the gap between cell and human studies. Each organoid typically begins with stem or progenitor cells from patients. In Engle’s lab, donor pancreas tissues were used to create miniature replicas of the pancreas. Findings based on a patient’s personalized organoid model could improve therapeutic effectiveness.
“By growing organoids directly from patients, we preserve key features of ductal cells and ask which disease mechanisms are active in each individual patient,” said Victoria Osorio-Vasquez, PhD, a postdoctoral researcher in Engle’s lab and first author of the study.
The researchers surveyed the molecular signatures in each organoid and found three subtypes of chronic pancreatitis. This biology-based patient stratification can inform optimal treatment.Results showed that approximately half of the organoids demonstrated dysfunctional CFTR.
“And CFTR dysfunction was not limited to patients with inherited CFTR mutations, suggesting that functional testing may identify therapeutic opportunities that would be missed by genetic testing alone,” Osorio-Vasquez says.
Existing CFTR modulator therapies treat patients with cystic fibrosis. The findings suggest that these same therapies may offer pancreatic benefits. The researchers tested clinically available CFTR modulators and found that these therapies could stabilize or restore CFTR function and reduce inflammatory signaling in responsive pancreas organoids.
The platform also revealed rare alterations to genes, KRAS and TP53, in some chronic pancreatitis organoids, supporting future use of the system to study disease evolution, pancreatic cancer risk, and biomarker discovery at the interface of chronic inflammation and pancreatic cancer.
“These organoids gave us a way to study chronic pancreatitis pathogenesis in human cells for the first time,” says Engle. “Our platform enables a more personalized way of studying and eventually treating chronic pancreatitis, while also blazing the trail for other organoid-based platforms in other inflammatory disease contexts.”
Adenosine deaminases acting on RNA (ADARs) have not been previously exploited for DNA adenine base editing in mammals. We developed a single-nucleotide-resolution adenine base editor (snuABE) that uses ADAR and customized guide RNAs to precisely convert adenines to guanines on DNA targets with minimal bystander editing.
AstraZeneca has acquired exclusive global rights to develop and commercialize Dizal Pharmaceutical’s marketed lung cancer drug Zegfrovy® (sunvozertinib), through an agreement that could generate up to $1.5 billion for the spinout of the pharma giant’s onetime Chinese R&D operation.
Wuxi City, China-based Dizal has inked an exclusive license agreement with AstraZeneca to expand the development of Zegfrovy into new indications beyond the one for which it has approvals in the United States and China—namely the treatment of adult patients with locally advanced or metastatic non-small cell lung cancer (NSCLC) with EGFR exon 20 insertion mutations, whose disease has progressed on or after platinum-based chemotherapy.
Dizal has been pursuing approvals from the FDA and China’s Center for Drug Evaluation (CDE) for a new indication for Zegfrovy, as a first-line treatment for NSCLC with exon 20 insertion EGFR mutations. In May, Dizal presented at the 2026 American Society of Clinical Oncology (ASCO) Annual Meeting and simultaneously published in The New England Journal of Medicine (NEJM), positive results in the indication from its Phase III WU-KONG28 trial (NCT05668988).
In the study, Zegfrovy showed a median progression-free survival (PFS) of 10.3 months compared with 7.5 months PFS for platinum-doublet chemotherapy in untreated NSCLC patients with EGFR exon 20 insertion mutations (exon20ins).
“The efficacy of sunvozertinib was superior to that of chemotherapy as first-line treatment for advanced NSCLC with EGFR exon 20 insertions,” the researchers concluded in their study, “First-Line Sunvozertinib in NSCLC with EGFR Exon 20 Insertion Mutations,” which was published May 29 in NEJM.
Dizal also showed Zegfrovy delivering a BICR-assessed best objective response rate (BoR) of 68.1% vs. 35.4% with chemotherapy, and a median duration of response (DoR) of 11.2 months vs. 7.1 months for chemo.
Based on those results, Dizal has filed supplemental New Drug Applications (NDAs) for Zegfrovy in the first line to the FDA and China’s Center for Drug Evaluation (CDE). Both regulators have granted their Breakthrough Therapy designations to Zegfrovy in that setting.
“AstraZeneca is a leader in treating EGFR-mutated lung cancer, and we are eager to add Zegfrovy to our world-class portfolio of innovative medicines for patients whose tumors carry exon 20 insertion mutations,” Dave Fredrickson, executive vice president of AstraZeneca’s Oncology Hematology Business Unit, said in a statement. “With this agreement, we will bring a differentiated, oral targeted treatment to these patients with limited options across the globe.”
20% jump
Dizal shareholders reacted to the agreement with AstraZeneca warmly enough to send shares traded on the Shanghai Stock Exchange jumping 20%, from RMB 46.94 ($6.93) to RMB 56.33 ($8.31). But the news did not appear to wow AstraZeneca investors, as shares of the pharma giant traded on the London Stock Exchange dipped 1.95% today, from 12,610 pence to 12,364 pence. Shares traded on the New York Stock Exchange also fell 1.95% as of 2:17 pm ET, from $169.47 to $166.16.
Dizal was established in 2017 as a joint venture between AstraZeneca and China’s State Development & Investment Corp. (SDIC), with AstraZeneca spinning out the R&D operations of its China Commercial Innovation Center to Dizal as well as three preclinical candidates, one each in cardiometabolic disease, respiratory disease, and oncology, the drug that was eventually developed into Zegfrovy. Xiaolin Zhang, PhD, who headed the innovation center, was appointed Dizal’s CEO, a position he still holds.
AstraZeneca has agreed to pay Dizal $600 million upfront; up to $900 million tied to achieving development, regulatory, and sales-related milestones; plus tiered double-digit royalties on global sales of Zegfrovy. The milestone payments consist of up to $400 million in clinical development-related payments and up to $500 million in sales-related payments, Dizal disclosed in a regulatory filing to the Shanghai Stock Exchange.
In March, Dizal reported that Zegfrovy generated about RMB 576 million (about $85.057 million) in revenue last year, up 85% from 2024. Zegfrovy accounted for nearly three-fourths (72%) of Dizal’s total 2025 sales of RMB 801 million ($118.282 million).
Zegfrovy is a once-daily oral irreversible epidermal growth factor receptor (EGFR) inhibitor approved by the FDA in July 2025 based on evidence from the Phase I/II WU-KONG1B trial (NCT03974022) in patients with locally advanced or metastatic NSCLC with EGFR exon 20 insertion mutations whose disease has progressed on platinum-based chemotherapy and received Zegfrovy 200 mg once daily with food.
WU-KONG1B enrolled 202 patients with locally advanced or metastatic NSCLC with EGFR exon 20 insertion mutations who had received previous platinum-based chemotherapy. The trial was conducted at 89 sites in the United States, Argentina, Australia, Canada, China, Chile, France, Italy, Malaysia, South Korea, Spain, and Taiwan.
AstraZeneca’s licensing deal with Dizal is expected to close in the second half of this year, subject to customary closing conditions and regulatory clearances. AstraZeneca said the transaction does not impact its 2026 financial guidance to investors, which it reaffirmed on April 29 as calling for a mid-to-high single-digit percentage increase in total revenue, and a low double-digit increase in “core” earnings per share from primary ongoing business activities.
“As a leading global company with a strong lung cancer franchise, AstraZeneca will help ensure patients around the world can benefit from this innovation discovered by Dizal scientists in China,” stated Zhang.
Summer is peak tick season, and with it comes familiar threats like Lyme disease and Rocky Mountain spotted fever. But scientists say another group of tick‑borne pathogens is quietly gaining ground: nairoviruses, a diverse family of negative‑sense RNA viruses carried by ticks across Asia, Europe, Africa—and now the western United States. Several nairoviruses can infect humans, causing high fevers, severe headaches, and, in some cases, organ dysfunction. One member of the family, Crimean‑Congo hemorrhagic fever virus (CCHFV), is often fatal and considered a global‑level threat.
Nairoviruses are found in ticks that feed on wildlife, livestock, and people. Recent human infections have been documented in China and Japan, including Songling virus (SGLV), Tacheng tick virus 1 (TTV1), and Yezo virus (YEZV). A related virus, Beiji virus (BJNV), caused an outbreak involving more than 100 patients in northeastern China. And on the U.S. West Coast, researchers recently identified Pacific Coast Tick nairovirus (PCTNV) in Dermacentor occidentalis, a tick already known to transmit Rocky Mountain spotted fever. As the paper noted, “Pacific Coast Tick nairovirus… was recently identified in Mendocino, California, from tick species known to harbor human pathogens and having a large presence across the state.”
The new study, titled “Insights into the Structure and Function of the OTU Protease Virulence Factors from Emerging Human Nairoviruses,” was published in ACS Infectious Diseases and reveals how these emerging viruses may slip past human immune defenses. All orthonairoviruses encode a specialized enzyme called ovarian tumor protease (OTU), which can remove small protein tags—ubiquitin and ISG15—from human proteins. Those tags normally act as alarm signals that activate immune responses, and removing them effectively evades the immune system. As the paper explained, “OTUs exhibit varying levels of deubiquitinating (DUB) and deISGylating activities that facilitate viral immune evasion, establishing them as key virulence factors.”
In the work, researchers isolated OTU proteases from four emerging nairoviruses—SGLV, TTV1, YEZV, and PCTNV—and compared their ability to strip immune‑signaling proteins. The standout was PCTNV, whose enzyme showed the strongest ability to remove both ubiquitin and ISG15. That suggests PCTNV may be unusually adept at evading human immunity, raising concerns because the virus is carried by a human‑biting tick common along the Pacific Coast.
The team also resolved high‑resolution crystal structures of several OTU proteases. These structural insights allowed the researchers to train computational models that begin to predict which nairoviruses may pose the greatest threat. “The biochemical and structural insights provide a path forward for predicting OTU activity among current and emerging nairoviruses,” the authors wrote.
Such predictive tools could help public‑health agencies monitor new tick‑borne viruses before they spread widely. As corresponding author Scott D. Pegan, PhD, of the University of California, Riverside, noted, “This study reinforces the need to be vigilant about not just tick bites but the type of ticks that an individual has been bitten by, as they may carry diseases beyond what we have been used to looking for.”
“I have this idea of a communal brain.” David Baker, PhD, told me as I sat in his office at the University of Washington (UW) surrounded by colorful and complex figurines of protein structures. It was the one-year anniversary of his Nobel Prize in Chemistry win.
Just outside his doors, a lab of more than 100 researchers was united by the shared ambition to design proteins from scratch (or de novo) for powerful applications across pharmaceuticals, vaccines, biosensors, and more. This “communal brain” housed at the UW Institute for Protein Design (IPD), where Baker led as director, was hard at work developing deep learning methods that could achieve atomic precision.
A small protein composed of 100 amino acids had an astronomical 20¹⁰⁰ possible sequences. Yet, only a vanishingly tiny fraction could fold into stable, functional structures.Misplacing a residue by an angstromcould mean the difference between a drug binding tightly to its target or complete failure.
For the antibody drug market worth hundreds of billions of dollars, Nathaniel Bennett, PhD, former postdoctoral researcher in the Baker lab, says AI-guided antibody design that bypasses the need for time-consuming experimental screens has long been a “holy grail” for a breadth of indications, including cancer and autoimmune disease.
Last November, Bennett and colleagues published a Nature paper demonstrating that full length de novo antibodies could bind user-specified epitopes. AI models could now construct antibody loops, the key region involved in binding that has been historically challenging to design due to its flexible nature.
Despite this technological leap, AI-designed proteins that were manufacturable, remained stable in the body, and avoided unwanted side effects, were still a step away. The gap fueled an industry debate over whethergenerating de novo medicines was even possible.
When I asked Baker to separate the hype from reality, he didn’t hesitate.
“The reality is that we can now design proteins on a computer,” Baker explained in our video interview.“The hype is that for therapeutics, there’s a lot more than the basic activity of a protein binding or catalyzing a reaction. Whether de novo proteins will revolutionize medicine will require improving our understanding of the biology.”
Nobel guests
Bennett is continuing molecular design research as a co-founder at Xaira Therapeutics. The AI-focused biotech launched in 2024 with over $1 billion in total funding and a star-studded leadership team, including Baker, as a scientific advisor, and Marc Tessier-Lavigne, PhD, former president of Stanford and CSO of Genentech, as CEO. Carolyn Bertozzi, PhD, Nobel laureate in chemistry, Scott Gottlieb, MD, former FDA head, and Alex Gorsky, former CEO of Johnson & Johnson,are among the board of directors.
“Science all becomes obsolete quickly because the field’s moving!” Baker told me. “The people that you mentor are more important than any science you do. They all go on and do great things.”
2024 Nobel Week was a testament to Baker’s scientific reach. Nearly 200 current and former members of his lab gathered in the Grand Hôtel in Stockholm to celebrate the newly named laureate, who was among a cohort of renowned AI experts who swept the awards ceremony.
Baker shared the Nobel Prize in Chemistry with Google DeepMind duo, CEO Demis Hassabis, PhD, and then-senior research scientist, John Jumper, PhD, whose AI model, AlphaFold, solved the protein structure prediction problem and has become one of the most widely adopted computational tools for drug discovery.
Meanwhile, the Nobel Prize in Physics was jointly awarded to Geoffrey Hinton, PhD, professor emeritus at University of Toronto, and John Hopfield, PhD, professor emeritus at Princeton University, for foundational discoveries that enabled machine learning with neural networks.
Together, the prizes represented a pivotal moment. AI was no longer confined to computer science but had become a transformative force across disciplines, earning recognition as a breakthrough deemed to confer the “greatest benefit to humankind.”
Back at the IPD, Baker’s research group spanned multiple floors. Yet, he knew everyone’s name, where they sat, and moved easily between conversations, bringing together researchers whose expertise might unlock a new direction. In the weeks after receiving the historic Nobel call, Baker chose to remain fully present for his team, implementing a strict “no travel rule,” despite the avalanche of invitations and media attention that accompanied the prize.
“David’s really good at forcing you to break the ice with people,” said Seth Woodbury, a graduate student who is designing metallohydrolases, enzymes that cleave some of the strongest bonds in biology for sustainability applications, including degrading pollutants. “Once you talk to your colleagues at happy hour, it’s not so scary to go ask them a question.”
Woody Ahern, graduate student and co-author of the metallohydrolase Nature paper, adds that Baker has a “very reasonable disdain for hierarchy.”
“Anyone can speak up in meetings. Anyone can question the work. It breeds this culture of staying focused on what matters in an interdisciplinary way,” said Ahern.
When Ria Sonigra was applying to graduate schools in the U.S., every option felt equally far from her home in India. She recalled sending Baker a cold email with questions about the lab. He quickly replied and offered to connect her with another international student who could help her navigate the application process. Today, Sonigra is an IPD graduate student, designing programmable nanopores for molecular sensing and sequencing.
“People outside the lab may think that David can’t pay attention to everyone, which is not true,” Sonigra said. “He knows your project and what he expects of you before the next meeting, even if he has a hundred trainees.”
At one point, Baker waved me over with a smile. “You’re missing chocolate hour!” he said, inviting me to one of many small weekly rituals that embodied the collaborative culture he had built.
“My group was not the first to do protein design,” he said humbly.
The field’s early innings trace back to 1988, when William DeGrado, PhD, demonstrated that sequences not found in nature could achieve stable 3D folds.The work challenged the long-held belief that functional proteins could only arise through evolution.
Steps toward computational design came a decade later, when for the first time, an in silico predicted protein was experimentally validated to adopt a target structure. The work was published in a Science study led by Steve Mayo, PhD.
Baker, alongside then-postdoctoral researcher, Brian Kuhlman, PhD, went a step further in 2003, expanding the design scope to include flexible backbones that represented entirely new folds, making it possible to not only modify natural proteins, but to create new ones from scratch.
“The prize was given because protein design has so much promise now, and that reflects the work of the whole community,” Baker continued.
Today, Degrado, Mayo, and Kuhlman are continuing to advance structural biology as prominent faculty members across University of California, San Francisco (UCSF), California Institute for Technology, and University of North Carolina (UNC) Chapel Hill, respectively.
Top7 was the first protein created on a computer with a custom amino acid sequence that folds into a never-before-seen structure. When viewed at an angle, the transparent partition allows the two forms to become superimposed, illustrating the beauty of uniting sequence and structure. [UW Institute for Protein Design]
Decades before OpenAI co-founder, Andrej Karpathy, coined the term “vibe coding,” Baker’s team was writing a program in FORTRAN. NamedRosetta, the molecular modeling suite simulated proteins atom-by-atom based on biophysical properties, from hydrogen bonds to backbone torsion angles.By calculating free energy, Rosetta could estimate which sequences were most likely to achieve a desired structure: the lower the energy, the more stable the predicted fold.
Yet, a protein’s energy landscape is rugged, with countless local minima among an astronomical number of conformations. Success was rare. Researchers were searching for a single grain of sand across the desert.
Still, “Rosetta was impressive,” said Sierin Lim, PhD, an associate professor at Nanyang Technological University, who is among a group of researchers engineering self-assembling nanoscale containers, known as protein cages, for applications across drug discovery, imaging, and materials science. She recalled watching molecules move on her screen in Singapore in the early 2000s. At the time, Rosetta was the only program that could model proteins.
Over the next twenty years, Baker adamantly pushed Rosetta to be openly available, inviting collaborators to not only use the software, but to improve it.
PyRosetta, a user-friendly Python-based implementation developed by Johns Hopkins University researchers led by Jeffrey Gray, PhD, broadened Rosetta’s access for structural biologists without a strong computational background. Meanwhile, progress in generating high affinity and selective ligand binders and epitope scaffolds for vaccine development were bringing computational proteins closer to real-world medicines.
What started as a single lab project grew into the Rosetta Commons, an international collaboration spanning more than 100 laboratories.
“It was a great move making Rosetta open, seeing what it can do now,” Lim said.
Researchers from Google introduced the transformer, a neural network architecture that enabled machines to analyze entire sequences at once. By using a “self-attention” mechanism, AI models could now uncover patterns across massive datasets at unprecedented scale. Soon, large language models (LLMs) trained on internet-scale text could not only understand, but converse in eloquent dialogue with humans.
The generative AI era had begun.
While the rest of the world was captivated by chatbots, structural biologists were sitting on a treasure trove of biological data pristine for machine learning.
For over fifty years, researchers had painstakingly deposited hundreds of thousands of experimentally determined structures in the Protein Data Bank (PDB) for public use. This molecular atlas now offered AI a window into the rules of biology.
In 2020, Baker received a phone call from one of the organizers of the Critical Assessment of protein Structure Prediction (CASP) competition, the biannual experiment that assesses the field’s latest state-of-the-art models.
“The first thing he said was, ’David somebody has done amazingly well this year, and it isn’t you!’” Baker recalled during his Nobel banquet speech. “That was how I first learned about the work of Demis and John.”
Instead of relying on human-defined biophysical rules, AlphaFold quickly learned decades of biochemistry from the PDB, uncovering the hidden instructions governing an amino acid sequence to fold into its 3D shape. At CASP14, the model remarkably predicted structures that were indistinguishable from real-world proteins. Months of laboratory work turned into a computational task completed in minutes.
Hassabis was quick to translate the breakthrough into medicine, taking the helm of DeepMind’s drug discovery spinout, Isomorphic Labs, as CEO a year later.
Today, the company’s IsoDD (Isomorphic Labs Drug Design Engine) platform, expands the druggable landscape by probing previously inaccessible biology, including predicting induced-fit interactions, where proteins change shape upon ligand binding, and identifying hidden binding pockets for drug targeting.
Isomorphic was betting, not on single therapeutic assets, but on a general discovery engine applicable across any disease area. That vision has since secured major pharma partnerships with Novartis, Eli Lilly, and Johnson & Johnson.
“I’ve always believed the No.1 application of AI should be to improve human health,” wrote Hassabis on LinkedIn when announcing Isomorphic’s whopping $2.1 billion funding raise in May.
Diffusion evolution
Concurrently, Baker’s team began applying deep learning to de novo design, drawing inspiration from AI’s emerging ability to generate realistic images. These diffusion models could operate on atomic coordinates and create entirely new protein backbones. Designs were conditioned for desired structural and functional constraints, opening the door to programmable biology.
When Baker’s team presented de novo design model, RFdiffusion (RoseTTAFold diffusion), in Nature in 2023, Mohammed AlQuraishi, PhD, assistant professor of systems biology at Columbia University, described the advanceas “a really big deal.”
‘‘Prior to the ‘diffusion evolution’, the success rates were probably on the order of 1 to 10,000, if you’re lucky,’’ AlQuraishi told me shortly after RFdiffusion’s publication. ‘‘With diffusion models, the success rates are closer to the single percentages when you get into the laboratory. It’s a huge magnitude improvement of what it used to be.”
Donald Hilvert, PhD, professor emeritus at ETH Zurich, met Baker twenty years ago while working on enzyme design with Defense Advanced Research Projects Agency (DARPA). Traditional Rosetta methods would carve out binding pockets in existing proteins and install a new catalytic apparatus.
“But the activities were not very good,” Hilvert recalled. Designing catalysis, where success depended on precisely positioning chemical groups to stabilize fleeting transition states, proved far more difficult than engineering a stable protein fold.Rosetta struggled to achieve that level of accuracy, prompting much of the field, including Baker, to turn attention elsewhere.
“Two years ago, David called me and said, ‘Why don’t you come and visit? All these new AI-driven techniques are really changing the game!’” Hilvert told me.
Hilvert has spent the past two summers at the IPD, collaborating with Woodbury, Ahern, and IPD postdoctoral researcher, Donghyo Kim, PhD, to design metallohydrolases using RFdiffusion. He “hardly knew how to turn on a computer,” yet was reading Python scripts and generating his first computational designs within weeks. To his amazement, experiments quickly yielded five or six promising hits.
“There is this common purpose of people helping one another,” Hilvert said. “David sets the tone from the top.”
Application generalist
As I walked through the halls of the IPD, I saw the extraordinary reach of protein design applications firsthand. Desks were intermingled across fields. The proximity was deliberate for ideas to travel as far as possible.
Florence Hardy, PhD, is a postdoctoral researcher tackling a new enzyme design project for global health applications, including streamlining the manufacturing process for therapeutics.
“I always say that I can only think in a ten angstrom sphere at a time,” she chuckled. “That’s just as big as the active site.”
“Most medicines focus on inhibitors,” Xinru Wang, PhD, explained when describing her postdoctoral research developing insulin agonists, or binders that lead to activation, to address metabolic disease. In contrast to blocking activity, “turning on” a signaling complex required precise structural tuning that was a natural fit for the IPD’s expertise.
Last November, Wang and colleagues published a study in Molecular Cell, demonstrating that de novo designed insulin receptor (IR) agonists could extend glucose-lowering effects. The findings offered a therapeutic alternative to escalating insulin doses, which is a known contributor to resistance. Notably, these engineered agonists avoided triggering cancer proliferation that is often associated with excessive insulin activation. Wang is currently an assistant professor at Northeastern University.
Tabitha Tcheau designs DNA binding proteins inducible with small molecules that can recognize novel pathogens and trigger the plant immune system. The highlight of her project, she says, is the ability to span interdisciplinary subgroups, from conformational dynamics, small molecules, and nucleic acids.
“One thing that blew me away here is that people are extremely supportive,” Tcheau told me. “Everyone you ask is super eager to help.”
Enisha Sehgal is among a team of researchers designing sequence specific DNA binding proteins that can power programmable transcription factors, targeted gene regulation, and new genome engineering tools.
“Being in this lab allows you to be a specialist in protein design, but a generalist in all the applications,” Sehgal said. “You get answers faster. You can iterate faster. Science moves faster.”
Visiting researcher and machine learning scientist, Kieran Didi, reiterates how the IPD’s interdisciplinary team enables rapid experimental validation of models. “I’m not going to spend two months in this fantasy world of computational benchmarks,” he said. “In the next week, I know if the model is actually working. Someone will quickly put it to the reality test.”
Postdoctoral researcher and chemist, Declan Evans, PhD, concurs and sees himself as the Alpha tester.
“I can go straight to the developer and say, ‘this is not how computational chemists would use this software,’” Evans said. “You can see changes being made in real time.”
Back in Baker’s office, he told me about his regular weekend escape to the mountains, one of the benefits of living in Seattle. Skiing and hiking were activities he valued highly. When asked to contribute an item to the Nobel Prize Museum, Baker chose a broken ski pole as a symbol that progress often comes through overcoming setbacks.
“But I don’t think people get ideas on top of mountains,” Baker tempered. “If you’re going to be a [principal investigator], you have to really like mentoring. For me, it’s super fun!”
Baker’s most enduring creation may not be any single protein, but rather the network he built—the diverse, inviting, and interconnected communal brain.
![Top7 was the first protein created on a computer with a custom amino acid sequence that folds into a never-before-seen structure. When viewed at an angle, the transparent partition allows the two forms to become superimposed, illustrating the beauty of uniting sequence and structure. [UW Institute for Protein Design]](https://www.genengnews.com/wp-content/uploads/2026/07/3D-printed-proteins-IPD-04.jpg)