Tom Miller, PhD, Iambic Therapeutics co-founder and CEO

Iambic Therapeutics has advanced its first pipeline candidate into the clinic, an AI-designed treatment for human epidermal growth factor receptor 2 (HER2)-driven cancers that moved from discovery to a first-in-human study in under two years—less than one-third of the average six-year conventional early drug development timeframe.

IAM1363 is a selective and brain-penetrant small molecule inhibitor of HER2 signaling for solid tumors. Iambic has dosed its first patient in a Phase I/Ib trial (NCT06253871) designed to evaluate the tolerability, pharmacokinetics, pharmacodynamics, and preliminary efficacy of IAM1363, both as monotherapy and in combination with trastuzumab in patients with advanced HER2 cancers. Trastuzumab is a HER2/neu receptor antagonist marketed as Herceptin® by Roche and its Genentech subsidiary.

The open-label, multi-center, dose escalation and dose optimization trial has an estimated primary completion date of April 2028.

In preclinical studies, according to Iambic, IAM1363 has shown over 1,000-fold selectivity for HER2 compared to EGFR, as well as a promising pharmacokinetic and safety profile, preferential tumor enrichment, and penetration of the central nervous system. In HER2 tumor models, including intracranial tumor models, IAM1363 demonstrated favorable efficacy and tolerability compared to benchmark tyrosine kinase inhibitors and HER2-targeted antibody-drug conjugates.

Iambic plans to present additional preclinical data for IAM1363 next week at the American Association for Cancer Research (AACR) Annual Meeting, set for April 5–10 in San Diego. That data, the company said last week, will be presented through a poster highlighting how the drug’s binding mechanism and potent HER2 activity overcomes multiple resistance mechanisms and how its EGFR avoidance is strong enough to enhance its safety.

By inhibiting HER2 wildtype and oncogenic mutant proteins, IAM1363 is designed to expand the therapeutic index—a ratio comparing the blood concentration at which a drug becomes toxic and the concentration at which it is effective—compared to available HER2 inhibitors, as well as avoid toxicities from off-target inhibition of the epidermal growth factor receptor (EGFR), a related receptor tyrosine kinase.

“Huge opportunity”

“We saw a huge unmet need and a huge opportunity associated with three things,” Tom Miller, PhD, Iambic’s co-founder and CEO, told GEN Edge. “One was expanding the efficacy versus tolerability with a more selective compound. Another was addressing more patients, including resistance mechanisms, by not only hitting HER2 wild type, but also the HER2 mutations that are disease-causing in many different areas of cancer. Then, third, being able to address the large number of patients that sadly develop brain metastases, which is a leading cause of morbidity in HER2-driven cancers.”

“We saw a very meaningful opportunity for a better drug to address those challenges, and we have one,” Miller added.

According to Iambic, IAM1363 engages the HER2 protein in a structurally distinct way from any previously known HER2 tyrosine kinase inhibitor (TKI) since it’s a type 2 TKI; all previously reported HER2 small molecule inhibitors are type 1 inhibitors.

“it’s a distinct scaffold. It leads to a novel structural engagement, and it leads to improved properties, including 10 times better selectivity for HER2 engagement versus off target tox than any other known TKI, 10 times better brain penetrance than tucatinib, as well as addressing all of those different additional mutations.

Tucatinib, marketed by Seagen as Tukysa®, is a HER2 brain penetrant TKI approved for indications that include treatment of adults with advanced unresectable or metastatic HER2-positive breast cancer (in combination with trastuzumab and capecitabine) who have received one or more prior anti-HER2-based regimens in the metastatic setting.

IAM1363 is Iambic’s first clinical candidate to be developed through its NeuralPLexer drug discovery platform, designed to identify therapeutic candidates with differentiated drug profiles by unifying physics-based machine learning and experimental automation.

Leading platform

NeuralPLexer gained attention in February when a study published in Nature Machine Intelligence showed the platform to have outperformed other platforms in applying generative AI to predict the structure of protein-ligand complexes as well as how these structures confirm after interacting with drug molecules.

“Owing to its specificity in sampling both ligand-free-state and ligand-bound-state ensembles, NeuralPLexer consistently outperforms AlphaFold2 in terms of global protein structure accuracy on both representative structure pairs with large conformational changes and recently determined ligand-binding proteins.

The same day that Nature Machine Intelligence spotlighted NeuralPLexer, Iambic announced publication of a white paper that introduced NeuralPLexer2, an improved version of the AI platform. NeuralPLexer2 features improved prediction accuracy for novel targets, an expanded scope of the model to include most categories of biological structures, and the addition of protein-protein complexes, cofactors, post-translational modifications (PTMs), and protein-nucleic acid complexes.

The expansion of capabilities encompasses nearly two-thirds (62%) of the 217,705 released structures in the Protein Data Bank (PDB). Some 134,000 refined structures from the PDB are used for training NeuralPLexer2, nearly doubling the training data compared to the original NeuralPLexer model, Iambic said.

NeuralPLexer2 also showed itself capable of predicting G protein-coupled receptors (GPCRs). By solely taking their protein sequences and ligand chemical structures as inputs on 32 recently determined GPCR structures, NeuralPLexer2 exhibited a median TM-score of 0.91—a level of accuracy that nears that of Nuclear Magnetic Resonance (NMR) experiments—compared with 0.75 for the first version of NeuralPLexer.

Fred Manby, PhD, DPhil, Iambic Therapeutics co-founder and Chief Technology Officer.

“That is a key piece of evidence that we use to support this statement, that NeuralPLexer is really shifting what it means to be structurally enabled as you execute a drug discovery program,” said Fred Manby, PhD, DPhil, who co-founded Iambic with Miller and is the company’s chief technology officer.

Pipeline of Innovation

“NeuralPLexer2 constitutes a very significant step-change in the performance of the model. That’s a continuous pipeline of innovation. We’re not stopping at NeuralPLexer2. We’re not,” Manby emphasized. “We’re also not making specific statements about future versions. But it’s an active area of research. And we’re very proactively making efforts in the direction of future NeuralPLexer versions.”

The Nature Machine Intelligence paper was written by researchers from Iambic, Caltech, and Nvidia. The Silicon Valley-based microprocessing giant, which has grown its market-leading footprint in AI chips to industries that include the life sciences, partnered with Iambic to develop NeuralPLexer through a collaboration launched last year.

Iambic uses that include NVIDIA A100 Tensor Core GPU and NVIDIA A10 Tensor Core GPU for model training, fine-tuning and inference; as well as the NVIDIA BioNeMo™ generative AI platform for drug discovery. Iambic also plans to access additional NVIDIA hardware, including the NVIDIA DGX Cloud AI supercomputing platform.

That collaboration arose from a collaboration during Miller’s days as a professor of chemistry at California Institute of Technology (Caltech) with colleague Animashree (Anima) Anandkumar, PhD, Bren Professor of Computing and Mathematical Sciences, who was then also a director of machine learning at Nvidia.

“That was a strong initial relationship with Nvidia,” Miller observed. “But the important thing is, over. The next three years, we have had several publications together. We have developed multiple technologies in that collaboration, It has been a rich and highly productive relationship, not only on the algorithmic design side, but also really deploying itself at scale and making sure that we can maximize the impact and use of these methods for drug development.”

The collaboration resulted in not just NeuralPLexer, but also OrbNet, the AI-accelerated machine learning applying quantum features and a graph neural-network architecture, which according to Iambic is 1,000 times faster than conventional quantum chemistry methods such as density functional theory (DFT), without compromising accuracy.

Nvidia gave Iambic a shoutout during its recent NVIDIA GTC 2024 AI developers conference, announcing that Iambic agreed to contribute its NeuralPLexer model as a BioNeMo cloud application programming interface (API) for noncommercial use, by helping researchers predict how a protein’s 3D structure changes in response to a drug molecule.

“We have a fantastic partnership with Iambic,” declared Kimberly Powell, Nvidia’s general manager and vice president, healthcare and life sciences, at GTC 2024. She cited NeuralPLexer as an example of generative AI methods and applications “that are impacting literally every single part of the discovery process.”

$100M financing

Nvidia is among investors that raised a combined $100 million for Iambic through an oversubscribed Series B financing completed in October. Nvidia joined three other new investors—Illumina’s venture financing arm Illumina Ventures, Gradiant Corp., and independent board member Bill Rastetter. They participated in the financing along with several existing investors, including Nexus Ventures, Catalio Capital Management, Coatue, FreeFlow, OrbiMed, and Sequoia Capital.

Iambic also took on its current branding, changing its name from Entos to better reflect its AI focus.

“We wanted a name that reflected the fact that we were a therapeutics company, and Entos did not have that in its name. So Iambic Therapeutics actually, explicitly uses that ‘therapeutics’ wording in it,” Miller said. “Iambic is, of course, a reference to the human language and we wanted to make sure that we retained the role of AI as a core identity of the company, even as we emphasized its therapeutic nature. So many of the AI advances are connected to the transformer models which also underpin large language models, and Iambic is a nod to that human language role within AI.”

The $100 million financing and rebranding were completed nearly three years after Catalio and Coatue led a $53 million oversubscribed Series A round for the company, then known as Entos and co-founded two years earlier by Miller and Manby, who at the time was a professor of theoretical chemistry at University of Bristol.

“Fred and I had incredibly rewarding careers in academia, and just really loved that mode of doing science,” Miller recalled. “But academic research is often the first part of an exciting story. You create technology innovations. But you don’t see that through. In most cases academics isn’t the natural way in most cases to see that through to clinical delivery of those scientific innovations.”

“I think we had a real hunger to see the translational, real world impact of the AI technologies and computational technologies we were developing, and creating a company was the right way to see that closed loop between innovation and human impact, that we wanted to participate in,” Miller added.

In February, Iambic celebrated the opening of its new headquarters in La Jolla, CA. The company has between 60 and 65 full-time employees, about evenly divided between its software and AI teams, and its drug discovery and drug hunting scientists.

“Over the course of the next year or two, we are expanding in a couple of different axes, most notably on the clinical side, so the numbers will increase over the course of the current year,” Miller said. “But we’re trying to do that in a way that helps us to kind of retain our flexibility and our nimble, fast-moving pace. So, we will see growth, particularly on the clinical side in the current coming year. But we’re doing that in a way that keeps us lean.”

Previous articleAACR 2024: Insilico Medicine to Present Preclinical Data for AI-Generated Cancer Drugs
Next articleAACR Co-Chair Christina Curtis Shares Themes and Highlights from the Meeting