Featured on the cover of Nature Machine Intelligence, one of Iambic Therapeutics‘ AI platforms, known as NeuralPLexer, can accurately predict protein-ligand complex structures. The platform excels at predicting the complex 3D structures formed when proteins bind with drug-like molecules. In a benchmark study featured in the publication, NeuralPLexer outperformed other systems, including AlphaFold2, in its ability to predict the often-subtle conformational changes that occur upon drug binding.
“I think that AI is a killer app for chemistry,” said Tom Miller, Ph.D., CEO and co-founder of Iambic Therapeutics. “The design of molecules that can be good medicines, catalysts, organic light-emitting devices, or polymers — chemistry is all about optimization processes in complex environments.” For drug discovery, that can involve taking heterogeneous data, satisfying multiple criteria in the eventual product, and searching a huge amount of possible combinations of atoms to make the promising molecule of interest. It’s inherently data-intensive, which is why AI is such a natural fit. AI systems excel at taking diverse inputs, creating multimodal inputs and predicting multiple outputs for optimization.
This need to juggle multiple parameters is a defining challenge in drug discovery, as Fred Manby, Ph.D., Iambic’s chief technology officer, points out. “You have to be able to predict properties, not just one property like the affinity for the target, but maybe 20 different properties,” Manby explained.
ProPANE: Multi-parameter lead selection in drug Discovery
Tom Miller, Ph.D.
To address this, Iambic relies on ProPANE, its multi-parameter lead selection platform. ProPANE is a massively pre-trained graph neural network deployed across dozens of drug properties for lead optimization. It’s supported by automated training, uncertainty quantification, and explainability features. “We use ProPANE as our key workhorse for learning from experimental data,” Manby continued, “analyzing endpoints like solubility, distribution coefficient (logD), hepatocyte clearance, microsomal clearance, efflux, permeability—anything that contributes to a drug’s ADME profile.
From AI design to the clinic
This emphasis on a holistic, multi-parameter approach is evident in the company’s lead drug candidate, IAM1363, a highly selective HER2 inhibitor now undergoing phase 1 clinical trials. This molecule, designed to overcome the limitations of previous HER2-targeting drugs, exhibits high selectivity, effectively targets difficult-to-treat mutations and demonstrates a favorable safety profile — all hallmarks of the AI platform’s ability to identify and refine promising drug candidates. The drug was discovered in less than 9 months and progressed from program launch to IND submission in less than 24 months — about a third of the industry average.
Why Iambic and NVIDIA joined forces
Fred Manby, Ph.D.
The development of Iambic’s suite of AI-driven drug discovery platforms, including NeuralPLexer and ProPANE, was facilitated in part through a multi-year research collaboration with NVIDIA, a leading provider of AI hardware and software. This alliance emerged organically from shared research interests and a mutual recognition of AI’s potential to advance drug discovery.
“It started a long time ago, pretty organically,” Miller said. “One of our key collaborators at Caltech was my longtime colleague there, Anima Anandkumar, who also was the director of machine learning at NVIDIA.” Miller estimates that he has co-written at least half a dozen papers with her involving Iambic (previously Entos) and NVIDIA. “[NVIDIA has] have joined our Series B as major investors, which has been announced. That came with an additional partnership to deploy the NeuralPlex technology on the BioNeMo platform for internal use.”
NVIDIA, one of the most notable AI companies of late, has helped push the envelop of machine learning. “We have certainly benefited from deep, genuine collaboration with their team,” Miller said. The collaboration has “manifested in the improvement of many of our key technologies over the past years.”
AI as a medicinal chemist guide, not a replacement
Iambic’s suite of technologies extends beyond NeuralPLexer and ProPANE to include Magnet, a generative molecular design platform, and OrbNet, an AI-accelerated quantum chemistry tool. OrbNet is noteworthy for its ability to calculate protein-ligand binding energies 1000 times faster than conventional methods without sacrificing accuracy.
While AI tools are powerful, Iambic recognizes that AI alone cannot solve the inherent complexities of drug discovery. “As soon as you find a compound that has good affinity for your protein target, you’re immediately on to the next problem,” explains Iambic’s CTO, Fred Manby. “What does it do in a cell? How selective is it? What’s its metabolic profile? Does it hit any safety issues? What’s it going to be like in pharmacokinetics? Drug discovery is about conquering a series of challenges.”
While some might envision using AI to automate drug discovery, Iambic Therapeutics has a different perspective. “We have never taken that approach,” Miller said. “Just to hit on this recurring theme about the fact that drug discovery is a hard job — AI will not make a hard job an easy job. It will make a hard job marginally more navigable. And I think that is really the approach we take.”
In the end, Iambic measures success not by the sophistication of its AI tools, but by the tangible results it achieves. “Iambic has always recognized that the only thing that really matters at the end of the day is whether or not this adds up to better medicines for patients with a higher probability of success and faster timeline,” Miller explained. “That’s easy to say, but actually delivering on that and being able to draw a straight line between the tools we developed, how they enabled us to search space faster, how that has led to this development candidate, and how it is working in humans — that is a narrative that not many companies have been able to execute on,” he said.
This dedication to translating AI innovation into real-world patient benefit is what drives Iambic forward, Miller said. “We really held [that patient-focused approach] as a guiding light in terms of the choices we’re going to make in terms of prioritization, technology and tools.”
Filed Under: clinical trials, Drug Discovery, machine learning and AI
