BioNTech (BNTX) AI Day 2025 summary

AI Integration and Strategic Vision

• AI is fully integrated across R&D, business operations, and manufacturing, focusing on oncology, infectious diseases, and personalized medicine, leveraging global AI centers and proprietary supercomputing infrastructure.

• Deep learning and reinforcement learning drive advances in model efficiency, compute, and data utilization, enabling breakthroughs in genomics, protein design, and drug discovery.

• Proprietary supercomputing infrastructure (Kyber cluster) and custom ML software stack support large-scale model training, with Kyber delivering ~500 PetaFLOPS and AIChor orchestrating over 15,000 experiments per month.

• A holistic, modular approach unifies multi-modal data and AI tools for next-generation therapeutics and vaccines, aiming to build a future pharmaceutical company with a fully integrated, AI-driven platform.

• Pursues a multi-platform oncology approach with 16 clinical programs and over 20 ongoing Phase 2 or 3 trials, including major pharma partnerships.

Key Technological and Scientific Milestones

• NTv3, a foundational genomics model trained on over 150,000 species, achieves state-of-the-art performance in genome annotation, variant prediction, and generative DNA design, validated in lab experiments.

• AbBFN2, a multimodal antibody design model, enables rapid, multi-parameter optimization and humanization, reducing experimental cycles and mutation load, validated in lab settings.

• InstaNovo technology advances de novo peptide sequencing for cancer-specific target discovery, doubling peptide identification rates and increasing accuracy by 10-15%.

• AI-assisted nanoparticle design enables de novo protein assembly for vaccine scaffolds, validated by electron microscopy and functionalized with antigens.

• AI-guided TCR engineering pipeline achieves up to 50,000-fold affinity enhancement, unlocking new therapeutic modalities and demonstrating in vivo tumor control.

Data Strategy and Collaborative Innovation

• Combines large-scale open-source data with high-quality, in-house clinical and experimental datasets for model training and validation.

• Digital pathology tools and active learning reduce annotation effort and improve model performance, increasing pathologist efficiency fivefold and accelerating biomarker discovery.

• AI models learn from human expertise and multi-modal biological data, with ongoing projects to integrate human feedback into model training.

• Strategic focus remains on advancing oncology and infectious disease pipelines, with openness to external collaborations and broader applications in genomics and protein engineering.

• Strategic partnership with Bristol Myers Squibb to co-develop pumitamig, a next-generation bispecific antibody in registrational trials across 10+ indications.