Automating genetic analysis could speed diagnosis of rare diseases


When Shayla Haddock was born in 1997, her parents could immediately see that something was wrong. Their newborn had unusual facial features, shorter-than-normal limbs and club feet. Hearing tests showed she was deaf.

Finding the correct diagnosis for Shayla took many years, and almost didn’t happen at all. As I describe in a press release published this week, Shayla’s genome was analyzed four years ago without turning up a diagnosis, although her doctors continued to suspect she had a genetic disease. But they nearly missed uncovering the answer, as our story explains:

On Aug. 10, 2012 — only two weeks after Shayla’s doctors at Lucile Packard Children’s Hospital Stanford concluded that they could not match her genetic patterns and symptoms to a disease — a scientific report about a newly discovered link between a genetic defect and a rare disease was published that would have allowed them to diagnose her. But at the time, genetic-testing results were not routinely re-analyzed to take into account new knowledge. The family and doctors remained unaware that the answer was out there.

Last year, as part of a scientific study, Shayla’s parents agreed to have her genome re-analyzed. This time, Stanford computer scientists used new computational tools they had developed to compare Shayla’s gene sequences to the scientific literature. They found the 2012 scientific report and predicted that Shayla had a rare genetic disease called Wiedemann-Steiner syndrome, which her doctors confirmed.

The scientific study, published recently in Genetics in Medicine, describes how Stanford researchers are tackling the larger problem of automating the diagnosis of rare genetic diseases. Right now, it takes 20 to 40 hours of a highly trained scientist’s time to make each diagnosis, and the process doesn’t take into account the rapid generation of new knowledge about genetic diseases. In the era of precision health, we can do better than that:

“With each passing month, more of the world’s genetic diversity is represented in scientific databases, and each time more information is there, it’s easier to interpret the next thing you see,” said Jon Bernstein, MD, Shayla’s clinical geneticist at Packard Children’s and an author of the new report.

“The genome is ultimately a programming language,’ [senior study author] Gill Bejerano, PhD, said. ‘We really would like to use machine learning and other approaches to build computer systems that leave as little as possible work for the human expert. A computer is going to be weaker than a human at doing this, but we think we can take the process 80 to 90 percent of the way by computer and provide a huge time savings for the human in the loop.”

Via Scope

Photo of Shayla Haddock by Norbert von der Groeben


Leave a Reply

  • (will not be published)