Alzheimer’s Disease (AD) poses one of the most significant healthcare challenges of our era, impacting millions globally while straining healthcare systems and delaying effective therapeutic breakthroughs. A fundamental barrier to progress lies in deciphering the complex cellular and molecular landscape of AD to enable early detection, targeted interventions, and personalised treatment pathways.

Recognising this, the ENFIELD scGPT-AD initiative was launched to accelerate Alzheimer’s research through advanced Artificial Intelligence frameworks. Within this ambitious initiative, LIBRA AI Technologies contributed its expertise in generative AI and Machine Learning, developing specialised models to analyse single-cell RNA sequencing (scRNA-seq) data to uncover key biological insights.

Bridging Complexity with Generative AI

Traditional computational methods struggle to detect the subtle biological signals hidden in scRNA-seq data, making it hard to identify markers of disease progression and potential treatment targets in Alzheimer’s.

LIBRA addressed this gap by applying Generative Pre-trained Transformer (GPT) architectures, adapted to process biological sequence data, to identify previously hidden gene co-expression networks, cellular state transitions, and biomarkers associated with AD. This work aligns directly with LIBRA’s mission to deliver AI-powered solutions that advance precision medicine and accelerate discovery pipelines in complex disease areas.

Strategic Outcomes Delivered

Over the course of the project, the LIBRA AI team has successfully:

  • Curated and standardised 96 high-quality AD-specific scRNA-seq datasets, harmonising over 1.7 million single cells from AD and control brains, building a solid foundation for large-scale model training. 
  • Developed and trained two model configurations, including an extended vocabulary enriched with 2,000 Alzheimer’s-relevant genes, enhancing biological coverage and interpretability. 
  • Achieved notable improvements in predictive performance, increasing AD vs. Control classification accuracy from 77% to 83% and Braak stage classification accuracy from 65% to 83%, demonstrating the model’s capability to capture disease-relevant signals. 
  • Mapped 141 AD-associated gene clusters encompassing 550 genes, with brain region-specific activity patterns particularly strong in the entorhinal cortex and hippocampus—two regions known to be central to AD progression. 

Amplifying Impact Through Strategic Engagement

During the ENFIELD Innovation Workshop at Boeing’s headquarters in Madrid, LIBRA AI Technologies presented its development of the scGPT-AD model, an advanced generative AI tool designed to analyse single-cell RNA data and uncover gene interactions and expression patterns in Alzheimer’s Disease.

Our Senior Data Scientist, Theodoros Siozos, showcased this work using an engaging PechaKucha storytelling format, fostering insightful discussions with researchers and innovators from across Europe on how generative AI can accelerate biomedical research and precision medicine efforts in neurodegenerative diseases.

Charting the Path Ahead

Building on the momentum and outcomes of the ADPT initiative, LIBRA is poised to advance its leadership in AI-driven healthcare innovation by expanding the reach and application of its generative AI frameworks. As part of this trajectory, our team published a peer-reviewed paper that was presented at the IEEE International Conference on Bioinformatics and Bioengineering (BIBE 2025), where we also highlighted the broader scope of our work.

For more information, the paper is available through the IEEE digital library here.

Looking ahead, LIBRA will continue to enhance its AI models and work towards integrating them into research and clinical pipelines. By collaborating with pharmaceutical and biotechnology partners, LIBRA aims to accelerate drug discovery in Alzheimer’s Disease, ultimately supporting more effective and targeted treatments for patients worldwide.