Iambic Launches Enchant v3 to Advance AI-Driven End-to-End Drug Discovery
Iambic, a clinical-stage life science and technology company developing novel medicines through an artificial intelligence-driven platform, has unveiled Enchant v3, the latest generation of its multimodal transformer model designed to advance end-to-end drug discovery and development.
The company describes Enchant as the “brain” behind its molecular superintelligence platform. When combined with Iambic’s in-house high-throughput experimental chemistry and biology capabilities, the platform is designed to learn from and predict a broad range of properties associated with potential drug candidates and target compound profiles.
The launch of Enchant v3 represents a major expansion of the model’s scale, data inputs and biomedical capabilities. According to Iambic, the new version has been developed to help increase the probability of success for drug candidates while enabling the company and its partners to explore additional therapeutic areas and drug modalities.
The model incorporates 41 billion parameters and has been pretrained on 4.5 trillion tokens. It can work across more than 6,000 molecular properties and 16 biomedical modalities, creating a framework intended to connect information from early discovery research with data generated later in the drug development process.
Building a Multimodal AI Platform for Drug Discovery
Drug discovery involves analyzing large amounts of highly diverse information. Researchers may need to integrate molecular structures, biological assays, protein information, images, genomic and other multi-omics datasets, pharmacokinetic measurements, clinical trial results and scientific literature.
Traditionally, these data sources have often been handled within separate workflows. Information generated during preclinical research may not be directly connected to clinical development datasets, creating what Iambic describes as “data walls” between different stages of the drug development process.
Enchant was designed to address this challenge by ingesting multiple types of public and proprietary information and learning relationships across different biological and chemical datasets.
The model can process data including assays, text, images, multi-omics, biomolecular structures and sequences, clinical trial information and other forms of biomedical data.
By combining these modalities, Iambic aims to create a computational system capable of understanding drug candidates from multiple perspectives rather than evaluating individual properties in isolation.
The company believes this approach can help researchers make more informed decisions during drug discovery and development.
Enchant v3 Expands Model Scale
One of the defining characteristics of Enchant v3 is its increased scale.
The model contains 41 billion parameters, which Iambic says provides greater reasoning complexity and supports more sophisticated predictive capabilities. It has also been pretrained on 4.5 trillion tokens, providing a substantially larger information base for the model.
The company has been developing Enchant through multiple generations and says it has observed predictable scaling behavior across those iterations.
According to Iambic, as the model has increased in size and incorporated additional data and modalities, its predictive performance has also improved across a diverse set of preclinical and clinical endpoints.
Fred Manby, PhD, Co-Founder and Chief Technology Officer of Iambic, described Enchant v3 as an architectural advancement over Enchant v2.
Manby said the new capabilities are intended to be deployed across Iambic’s internal drug discovery and development programs as well as efforts conducted with partners. He also noted that earlier versions of Enchant have demonstrated utility across the company’s pipeline, including prediction of important preclinical and clinical endpoints.
More Than 6,000 Molecular Properties
Another important component of Enchant v3 is its ability to work across more than 6,000 molecular properties.
Drug development requires optimization across numerous characteristics. A molecule may need to demonstrate appropriate potency and selectivity while also possessing suitable pharmacokinetic, safety and other properties. Improving one characteristic can sometimes affect another, making multi-parameter optimization an important part of drug discovery.
Iambic designed Enchant v3 to address this complexity by simultaneously considering a large number of molecular properties.
The company believes this capability can help researchers evaluate potential compounds more comprehensively and identify candidates that have a more favorable combination of characteristics.
Instead of focusing on a single endpoint, the platform can use information from multiple related properties and biological measurements to guide decision-making.
16 Biomedical Modalities
Enchant v3 incorporates 16 biomedical modalities, expanding the range of information that can be processed by the model.
These modalities include text, assay data, molecular structures, protein structures, three-dimensional structures, images, multi-omics, biologics, clinical trial data and pharmacokinetics.
The integration of these different data types is intended to give the model a broader view of the chemistry and biology underlying potential drug candidates.
Importantly, Iambic is also using clinical development information within the platform. This is intended to help connect earlier-stage drug discovery with later-stage clinical considerations and potentially improve the ability to identify characteristics associated with clinical tractability.
The approach is designed to move beyond the traditional separation between preclinical discovery and clinical development.
Connecting Preclinical and Clinical Data
A major objective of Enchant is to break through data barriers that have historically separated different stages of pharmaceutical research.
Preclinical drug discovery generates extensive information about molecules, biological targets and experimental systems. Clinical development, meanwhile, produces information about patient populations, pharmacokinetics, safety, efficacy and treatment outcomes.
Connecting these datasets could potentially allow models to learn relationships between early molecular characteristics and later development outcomes.
Iambic says Enchant has been designed to use information across these stages rather than treating them as completely separate data environments.
This capability could be particularly important for improving candidate selection. A compound that performs well in an early laboratory assay may not necessarily have the properties required for successful development. Incorporating broader datasets may help researchers identify potential risks or opportunities earlier in the process.
Probability-Guided Decision-Making
Enchant also incorporates uncertainty quantification to convert model predictions into probability-guided decisions.
Rather than simply generating a prediction, the system is designed to provide information about the uncertainty associated with that prediction. Iambic believes this can help researchers determine which laboratory experiments are most valuable and where additional experimental data could provide the greatest amount of information.
This approach is intended to improve the efficiency of experimental drug discovery.
When thousands of potential compounds or molecular configurations could be evaluated, researchers must decide which experiments to conduct first. A model capable of identifying areas of uncertainty may help prioritize experiments that are most likely to distinguish between competing candidates or improve understanding of a molecule’s properties.
The resulting combination of AI prediction and experimental validation forms an important part of Iambic’s molecular superintelligence approach.
Scaling Laws Across Drug Discovery
Matt Welborn, PhD, Senior Vice President of Machine Learning at Iambic, highlighted the importance of the company’s experience with model scaling.
When Iambic introduced Enchant in 2024, it was not clear whether scaling principles observed in areas such as language modeling would translate effectively to drug discovery, according to Welborn.
Biomedical data is more complex and heterogeneous than conventional text datasets, but it also offers a wide range of information that can be integrated into increasingly capable models.
Welborn said Iambic has observed consistent improvements across three generations of Enchant. According to the company, adding parameters, data and modalities has resulted in improved predictions, including for endpoints where available data is limited.
This observation forms part of the rationale for Enchant v3’s substantially expanded architecture and training dataset.
Supporting Internal and Partner Programs
Iambic intends to deploy Enchant v3 across both its internal pipeline and drug discovery programs conducted with partners.
The company has identified two central objectives for the new version. The first is to increase the probability of success for drug candidates. The second is to expand the platform into additional therapeutic areas and modalities, allowing it to support a broader range of internal and external programs.
The company’s strategy combines AI capabilities with its own high-throughput experimental chemistry and biology infrastructure.
This integrated approach allows computational predictions to be connected with experimental work. Model predictions can inform which compounds should be synthesized or tested, while experimental results can subsequently provide additional data for improving model performance.
Such an iterative process is central to Iambic’s broader molecular superintelligence platform.
AI’s Expanding Role in Pharmaceutical R&D
The launch of Enchant v3 reflects the broader development of AI technologies for pharmaceutical research. Drug discovery requires researchers to navigate enormous chemical spaces while balancing multiple biological, pharmacological and development-related requirements.
Traditional approaches rely heavily on sequential experimentation and expert interpretation. AI models can potentially complement these methods by identifying patterns across large datasets and generating predictions for molecules or properties that have not yet been experimentally characterized.
Iambic’s approach is focused on applying these capabilities across the full drug development continuum rather than limiting AI to a single discovery task.
By combining chemical and biological data with clinical information, the company is attempting to develop a more integrated computational framework for drug development.
Advancing the Molecular Superintelligence Platform
Enchant v3 represents the latest stage in Iambic’s effort to build an AI-driven system capable of supporting multiple aspects of drug discovery and development.
With 41 billion parameters, 4.5 trillion training tokens, more than 6,000 molecular properties and 16 biomedical modalities, the new model substantially expands the scale and breadth of the platform.
The company’s emphasis on predictable scaling, uncertainty quantification and multimodal learning is designed to support better-informed experimental decisions while connecting data generated at different stages of pharmaceutical research.
Iambic plans to apply Enchant v3 to its own pipeline and partner programs, with the broader goal of improving the probability of success for potential medicines and extending the platform into new therapeutic areas and modalities.
As AI continues to evolve from a supporting analytical tool into an increasingly integrated component of pharmaceutical research, Enchant v3 represents Iambic’s latest effort to combine large-scale multimodal modeling with experimental chemistry and biology. The company expects the technology to play a central role in its ongoing drug discovery and development activities, potentially helping researchers evaluate complex molecular profiles, prioritize experiments and advance promising candidates toward clinical development.
About Iambic
Iambic’s mission is to make better technology for better medicines. By utilizing molecular superintelligence — integrating proprietary AI with automated, scalable chemistry and biology experimentation — our platform is designed to address the totality of drug discovery and development. We believe the utility of our platform is demonstrated by the discovery of a novel drug candidate advanced to clinic in approximately two years, along with a diverse preclinical pipeline. Our leading AI technologies include Enchant and NeuralPLexer for multimodal endpoint prediction designed to improve the speed, precision, and success rates of drug discovery and clinical development. At Iambic, we are building an engine that is designed to power the future of medicine. | iambic.ai