Evaxion Biotech A/S, a clinical-stage TechBio company specializing in developing AI-Immunology™ powered vaccines, launches an enhanced version of its clinically validated AI-Immunology™ platform with an update of its EDEN™ AI prediction model. Among other improvements, the model can now predict toxin antigens, allowing for the development of improved bacterial vaccines.
Bacterial toxins are often key contributors to disease, making their neutralization essential for developing effective vaccines. The upgraded EDEN™ model improves the prediction of bacterial toxins for vaccine development. With greater speed and accuracy, it enables Evaxion to rapidly identify the critical toxin targets to include in vaccines.
“Today’s launch of the improved EDEN™ AI prediction model marks an important milestone for Evaxion, further strengthening our AI-Immunology™ platform. As one of the few truly AI-first TechBio companies, our AI-Immunology™ platform is at the forefront of innovation. We will continue to invest in its development and refinement to further improve our ability to discover novel targets and develop advanced vaccines,” says Christian Kanstrup, CEO of Evaxion.
The AI-Immunology™ platform uses advanced AI and machine learning technologies to design and develop novel vaccine candidates addressing significant unmet needs. Its AI prediction models are applied in cancer and infectious diseases and scalable to other therapeutic areas. The platform can deliver one new target within just 24 hours compared to years by using traditional methods and is delivery modality agnostic. The predictive capabilities of the AI-Immunology™ platform are robustly validated as the target’s prediction score has been shown to correlate with pre-clinical and clinical readouts.
The EDEN™ prediction model is one of five models constituting the AI-Immunology™ platform. It is used to identify B-cell antigens included in infectious disease vaccines. The new version 5.0 features the following updates:
- Novel bacterial toxin antigen predictor: We have trained new machine learning models, improving the accuracy and reliability of toxin antigen prediction
- Expanded training dataset: We have streamlined the process for curating additional training data from published sources using retrieval-augmented generation with large language models, followed by manual domain expert curation
- Advanced protein feature prediction: We developed a new building block for protein feature prediction using protein language models, enhancing the model's architecture and capability to predict various protein characteristics
The data documenting the features and performance of the new EDEN™ prediction model will be presented today in a poster session at the European Conference on Computational Biology (ECCB) in Turku, Finland.