Exploring the power of AI in genomics

New AI-driven solutions promise to improve whole genome sequencing and diagnostics, offering insights into genetic diseases and complex conditions such as Alzheimer's and type 2 diabetes. In this interview with MEDICA-tradefair.com, Dr. Uirá Souto Melo explains the technology behind the work, the challenges it faces, and the vision for the future.

Could you briefly explain what Lucid Genomics is doing, particularly how your AI solutions differ from other whole genome sequencing technologies?

Dr. Uirá Souto Melo: Lucid Genomics is a project that started about five to six years ago. We developed a machine learning tool that enhances the quality of affordable, low-quality DNA sequencing data by learning from clean, high-quality data. Many companies still stick to sequence their patients with the affordable but noisy technology, and our AI improves this data to extract 100 percent of the genomic information. This approach not only helps diagnose genetic diseases more accurately, but also uncovers previously missed genomic signatures in complex diseases like osteoporosis and type 2 diabetes.

How does your technology handle data from different sequencing platforms, such as short reads and long reads?

Melo: Over the last six years, we've developed algorithms to process data from all types of sequencing technologies, including short reads and long reads. Our AI analyzes the raw data from these platforms and enhances it. We then present the results in an interactive software interface, allowing users to explore the data and detect previously hidden genomic signatures.

What are the challenges in interpreting non-coding mutations, and how does your technology address them?

Melo: Non-coding DNA is a relatively new field of study. Until recently, the sequencing technology wasn’t good enough to analyze non-coding regions, so researchers largely ignored them. These regions are crucial, as they control the expression of genes – like wall switches that turn lamps (genes) on/off. The big challenge is to map the "wiring" of these non-coding regions. We’ve developed AI models to predict which gene (lamp) will be affected (on/off) if we detect a mutation ("flip") in a certain switch (non-coding elements), helping us better understand the impact of non-coding mutations.

What role do structural variants play in diagnostics, and how does Lucid Genomics handle them?

Melo: Structural variants, which are larger mutations, are often missed because they lie outside the two percent of the genome typically sequenced. These variants are important for understanding many genetic diseases, but most diagnostic tools can’t detect them. We solve this by using AI to improve affordable sequencing technologies, allowing us to identify structural variants in the non-coding DNA that can cause disease. We’ve already demonstrated this in patients with heart, bone, and neurological diseases.

Are you planning to integrate 3D genome analysis and epigenetic data into your platform?

Melo: Yes, and this is what sets us apart. Our software not only extracts 100 percent of the genome data, but also integrates 3D genomics and epigenetic information. DNA isn’t just a linear sequence; it’s folded in specific ways inside the nucleus, and this folding affects gene expression. We use this 3D structure to map how genes are turned on and off. We’ve also developed machine learning tools to map the epigenetic landscape for specific tissues, such as the retina and heart, which further improves our ability to detect impactful variants.

Data security is a major concern nowadays. How do you address privacy and security issues with such sensitive genetic data?

Melo: Data privacy is indeed a top priority. Most of our clients, such as hospitals and research institutions, require our software to run on their premises to ensure data doesn’t leave their secure environment. For those using cloud services, we comply with the strict General Data Protection Regulation (GDPR) in Europe. We use Amazon Web Services (AWS) based in Frankfurt, and our head of technology has over 15 years of experience in cybersecurity to ensure the highest level of data protection.

Can you share a recent success story where your technology made a significant impact?

Melo: One case involved seven families from around the world, all affected by a rare heart condition called atrial fibrillation. Despite decades of study, the cause remained unknown. After collaborating with a French research group, we identified a structural variant in the non-coding DNA that altered the "wiring" of gene regulation, leading to the condition. This discovery was validated in a mouse model, which recapitulated the disease. In another case, our AI identified a previously unseen pattern in acute myeloid leukemia (AML) patients, which could lead to new treatment strategies.