March 19, 2026
In the last two decades, researchers have made substantial advances in genomics, enabling them to identify more than 7,000 rare genetic diseases. Despite this remarkable progress in genome technologies and variant interpretation pipelines, more than half of rare disease cases remain unsolved, largely because many gene-disease associations are still unknown or are extremely rare. As a result, many patients will wait years for a diagnosis. During this time, their symptoms can progress and their hope can start to fade. For people living with a rare disease, a timely and accurate diagnosis can be transformative—guiding care, reducing uncertainty, and offering peace of mind.
To address this challenge, Genome-wide Sequencing Ontario (GSO), a collaboration between The Hospital for Sick Children (SickKids) and the Children’s Hospital of Eastern Ontario (CHEO), is leveraging Illumina technology and expertise. GSO is funded by the Ontario Ministry of Health to provide clinical genome-wide sequencing services for Ontarians with a suspected rare disorder. Since launching in 2021, eligibility has expanded and demand for testing has increased substantially, resulting in the need for more efficient analysis while maintaining quality. “For patients with a rare disease, receiving a diagnosis can sometimes take years,” says Christian Marshall, a clinical molecular laboratory director in the Division of Genome Diagnostics at SickKids. “This is partially driven by our inability to interpret much of the rare variation in the genome and is compounded by bottlenecks in manual analysis and reporting. Improving variant annotation pipelines and using new tools will be essential to delivering accurate and timely results for patients.”
Evaluating AI to deliver faster, more accurate diagnoses
But improving these pipelines can be a time-consuming and labor-intensive process and doing so at scale with human analysts is not sustainable. The analysis workflow has multiple steps, from variant calling, which takes the patient’s genome and compares it to a reference genome to find variants, to variant annotation, which adds context to the variants using known databases. This information is then used to interpret a variant’s potential pathogenicity, a process that has traditionally been carried out manually.
In hopes of making a real impact on diagnostic yield and turnaround time, the team decided to evaluate the application of artificial intelligence (AI) on the variant analysis workflow. “It can be very difficult to understand if a variant is pathogenic, even with annotation software,” says Livia Loureiro, a senior staff medical science liaison in the Medical Affairs Department at Illumina. “A single genome contains roughly 4-5 million variants. Genome sequencing captures a broad spectrum of variation—including single nucleotide variants, structural variants, and short tandem repeats—making the data far richer, but also far more complex to interpret. Without adding AI to the workflow, analyzing data at that scale with speed and precision just wouldn’t be possible.”
Using secondary analysis and explainable AI variant interpretation software for tertiary analysis, the team conducted a retrospective study using 852 pediatric cases with known causative variants. They evaluated a new AI model and compared it to the previous version. They found that both AI versions were successful in prioritizing causative variants in patient cases, demonstrating that 98.3% were ranked in the top tier (“Most Likely”) in version one and 98.8% in version two. When considering all prioritized predictions, both models achieved 99% capture of causal variants. They also found that the new AI version showed improvement in specificity, picking up 100% of variants reported in the top ten, compared to the older version, which picked up 94% of variants.
“This is an important milestone,” says Loureiro. “The evidence supports the use of AI-driven variant analysis to prioritize variants both quickly and with unprecedented precision on a large number of cases.”
Marshall hopes that AI will be able to go a step further than variant interpretation and support the collection of structured phenotypes from clinical notes to further improve the accuracy of variant interpretation. Of course, there will still need to be some human oversight, particularly for variants with low penetrance or that have sequencing or mapping quality issues. “AI is poised to have a significant impact on diagnostics for rare diseases,” says Marshall. “But I also see the need for detailed human evaluation before it is fully integrated. The key, and also the main challenge, will be designing robust validation and using it responsibly.”
When data is shared, more answers are possible
To amplify the potential impact of AI-driven variant interpretation, Loureiro and Marshall are also developing a shared data network across laboratories in Canada. Because rare diseases affect so few people, clinicians may not have previously encountered a patient’s symptoms, and the gene–disease associations may not yet be captured in the databases used by genomic analysis software. Even comprehensive databases like ClinVar and gnomAD contain gaps when it comes to rare disease variants. This makes diagnosing rare diseases extremely challenging. Shared data networks enable laboratories to pool anonymized patient data and apply matching algorithms to find unrelated individuals with similar phenotypes or genetic variants. This approach breaks down clinical and geographical silos and helps clarify variants of unknown significance, enabling laboratory technicians to identify patterns that would otherwise remain hidden and improve the likelihood of a diagnosis.
“We’re already seeing the benefits of shared data networks for our patients,” says Marshall. “Even with just a few thousand samples, we’ve identified variants multiple times that didn’t appear in gnomAD, which is incredibly valuable for variant assessment. For example, if my patient has a rare variant that was identified in another patient within our network, but the phenotypes don’t match, then it helps rule it out. And if they do match, it helps us determine the diagnosis.” While data sharing offers clear benefits, it also raises important privacy challenges, such as information management, patient consent, and cross-institution data agreements. Addressing these considerations—and demonstrating clear clinical value—will be essential to scaling these networks.
Loureiro and Marshall believe AI-assisted variant analysis technologies and shared data networks have the potential to deliver more accurate and timely diagnoses, advance scientific research, and accelerate discovery for rare diseases—and ultimately ease the diagnostic journey for rare disease patients and their families.
