Annabelle was born five years too soon. Then a machine made a diagnosis

A remarkable automated system has diagnosed hundreds of people living with rare and unnamed conditions by repeatedly analysing huge troves of genetic data and a steady pipeline of new discoveries.

The open-source tool called Talos, designed by Australian and US researchers, has detected rare genetic diagnoses for more than 350 children and adults, opening up treatment pathways and providing families with crucial information to plan for the future.

“That number keeps going up and up,” said senior researcher Professor Zornitza Stark, clinical geneticist at Murdoch Children’s Research Institute (MCRI) and Victoria Clinical Genetics Services.

“It’s the gift that keeps on giving.”

About one in 17 Australians is affected by a rare disease, many of whom spend years searching for answers that may never have materialised. Genomic testing has revolutionised scientists’ ability to identify these conditions but about half of these patients remain undiagnosed after genomic testing.

Every year, hundreds of gene-disease associations and thousands of new ways to interpret them are published, said fellow senior researcher Professor Daniel MacArthur, director of the Centre for Population Genomics, a Garvan Institute and MCRI initiative.

Stark said it would be impossible for human analysts to repeatedly, manually reanalyse thousands of patients’ genetic data. The labour intensity and cost would be prohibitive.

“This means a lot of rare disease families still miss out on a diagnosis, even though it is sitting in the data,” Stark said.

The researchers designed Talos to automate this process and incorporate any newly published gene-disease associations, then alert them to potential new diagnoses it found.

To validate the tool, the researchers fed Talos Australian and US datasets comprising 1089 patients who had already undergone exhaustive genomic investigations by human analysts. Talos found up to 90 per cent of the diagnoses that the expert human analysts detected, with one key difference: it was faster and cheaper.

Then they had Talos analyse a database of 4735 children and adults with neurodevelopmental disorders, inherited heart or kidney conditions, who were still undiagnosed after initial genomic testing.

Talos detected 241 new diagnoses at the time the researchers submitted their paper to Nature Medicine.

The median time it took Talos to make a new diagnosis after the corresponding discovery was published was 32 days. Some diagnoses were made the same day, the research team, including scientists at Broad Institute of MIT and Harvard, and Microsoft Research, reported in the prestigious journal on Thursday.

Annabelle Holland was born five years before scientists identified the variant that caused her rare genetic condition.

Her mother, Keira Hannigan, had agonised over whether she had done something during her pregnancy to cause Annabelle’s profound intellectual disability, cerebral palsy, seizures and visual impairment.

“Had I drunk too much coffee? Was it some medication I took?” Hannigan asked herself.

As a newborn in intensive care, Annabelle underwent ultra-rapid genomic testing, to no avail. Her genomic information was kept on file until, in 2024, Talos detected a diagnosis.

ReNU syndrome had just been identified. The syndrome is caused by variants in the RNU4-2 gene, affecting brain development and function. Annabelle’s ReNU syndrome had occurred spontaneously, like the vast majority of cases.

“The relief I felt … it wasn’t my fault,” Hannigan said. “We were simply unlucky.”

The diagnosis meant Hannigan and her husband, Ben, could have another child, nine-month-old Victoria, knowing that it was extremely unlikely she would be affected and that they could test for the syndrome.

“It gave us answers among the uncertainty,” she said, “and connects you to an international community of families who truly understand what you’re going through.

“And the way medicine is moving, that gives us real hope.”

Stark said Talos could transform outcomes for patients and families, including treatment options.

More than 50 families who received a diagnosis had undergone further testing to inform decisions around screening and having more children.

It costs roughly $17 a month to run Talos, the researchers reported. Stark said other research groups can run the open-source tool as frequently as they like.

“We very much hope that many other diagnostic laboratories around the world will implement this.”

Stark said the same principles could be applied to identify rare cancer diagnoses, and her team was planning to incorporate AI technology to further improve Talos’ performance.

“Time is of the essence when it comes to quickly and accurately diagnosing rare cancers,” said Christine Cockburn, chief executive of Rare Cancers Australia.

An RCA survey of 2500 patients with rare and less common cancers found 21 per cent had at least a six-month delay in diagnosis. For 7 per cent it took more than three years.

“Automation and machine learning are wonderful opportunities to change outcomes by speeding things up,” Cockburn said.

“There is untapped data that really matters to human beings, and this is really exciting.”

Start the day with a summary of the day’s most important and interesting stories, analysis and insights. Sign up for our Morning Edition newsletter.