Eye2Gene: Revolutionising Diagnosis for Genetic Eye Disease

Eye2Gene is a pioneering UCL startup transforming how rare genetic eye diseases are diagnosed and treated—by combining AI, retinal imaging, and privacy-preserving federated learning.

Founded by Associate Professor Nikolas Pontikos at the UCL Institute of Ophthalmology, Eye2Gene joined cohort six of the CDI Impact Accelerator to tackle one of its biggest challenges: overcoming data-sharing barriers that limited the scalability and accuracy of its AI platform. 

By leveraging breakthroughs in medical imaging, and genomics, Eye2Gene’s AI-powered precision ophthalmology platform aims to accelerate diagnosis and support the development of treatments for rare genetic eye diseases. However, restricted access to global hospital data posed a serious limitation— making it difficult to improve diagnostic performance across diverse populations. 

With support from the CDI and UCL Advanced Research Computing (ARC), Eye2Gene partnered with Flower Labs to integrate federated learning into its system, enabling its AI model to be trained on retinal scans from hospitals worldwide without transferring sensitive patient data—expanding the platform’s global impact. 

The Challenge: Unlocking AI’s Potential Without Sacrificing Privacy 

Genetic eye conditions are the leading cause of blindness in children and working-age adults, affecting more than 10 million people globally. Yet fewer than half of these conditions can currently be genetically diagnosed, and testing remains largely inaccessible outside a handful of high-income countries. The lack of diverse data stalls progress—not just in diagnosis, but in research and treatment development too. 

AI has the potential to transform retinal imaging—advancing diagnosis, accelerating research, and enabling life-changing gene therapies. But real-world progress is hampered by the fact that medical data is often siloed within individual institutions. Global data protection laws further restrict access, making it difficult to train AI models on the broad datasets needed for reliable, generalisable results. 

“Such a scarcity of information severely hinders the development of much-needed therapies in this space that can prevent and restore sight loss.” says Nikolas Pontikos. 

The Solution: A Federated Future for Precision Ophthalmology

Eye2Gene’s platform applies AI to retinal imaging to improve genetic diagnoses by predicting the likely causative gene and to support the development of treatments by monitoring disease progression over time. 

Initially, the Eye2Gene model was trained on a centralised UK dataset, which limited its ability to recognise patterns across global populations. To scale without compromising patient privacy, the team needed a new approach. 

During the CDI Impact Accelerator, Eye2Gene partnered with federated learning experts Flower Labs to create a secure system for training its AI model on hospital data worldwide, without transferring any patient information. With technical support from AWS and the CDI, they built an infrastructure capable of running complex AI training across cloud environments in the UK, Germany, and Australia. 

Accelerating Growth with the CDI Impact Accelerator 

To demonstrate federated learning at scale, Eye2Gene launched pilot implementations at leading sites in Bonn, Germany, and Melbourne, Australia. 

The team set up segregated AWS Virtual Private Cloud (VPC) environments in each country, each hosting an Amazon EC2 instance and S3 storage for test data. Flower was deployed across these sites using demonstration code to validate the setup. Meanwhile, Eye2Gene’s training pipeline was converted into a Nextflow script and executed using AWS Batch via the Seqera platform in order to make use of AWS spot instances. 

A custom Flower script was then developed to coordinate model training and manage the flow of federated weights between Flower and Nextflow. This setup enabled Eye2Gene to offload intensive computation to on-demand AWS spot instances, training the model efficiently across multiple locations. 

Impact Beyond Rare Eye Diseases 

Eye2Gene’s integration of federated learning represents a groundbreaking technical achievement—the first known deployment of Flower federation for this specific application. This innovative approach combines scalability, privacy, and regulatory compliance in an increasingly complex healthcare landscape. 

While Eye2Gene’s initial focus is on rare eye diseases, the potential applications extend beyond ophthalmology. By establishing a flexible and robust federated learning framework, the team has laid the groundwork for broader adoption of AI across medicine. This addresses one of the biggest challenges in healthcare: enabling secure, cross-border data collaboration that accelerates research and clinical progress. 

Growing Momentum and Expanding Reach 

The momentum behind Eye2Gene is gaining significant traction. The platform has led to eight academic publications and a strategic partnership with Heidelberg Engineering, a world leader in retinal imaging technology. Eye2Gene was invited to present at the 2025 Flower AI Summit, and a manuscript has been recently accepted Nature Machine Intelligence, with special mention of CDI’s crucial role in its success. Interest continues to grow, with over 50 potential customers now engaged. 

In 2025, the startup secured two key grants—one from Sight Research UK and another from the Medical Research Foundation—funding the launch of its first NHS clinic pilot and the deployment of Eye2Gene across multiple global sites. This expansion aims to screen patients for treatment eligibility and monitor a global cohort receiving gene therapy. 

To learn more about Eye2Gene please visit the website: www.eye2gene.com