A.I. Speeds Infant Genetic Disease Diagnostics

(EcoBear, Flickr)

25 Apr. 2019. Hospital, academic, and industry researchers devised automated techniques including artificial intelligence to sharply reduce the time for diagnosing genetic diseases in infants. Results of the study conducted at Rady Children’s Institute for Genomic Medicine in San Diego appear in yesterday’s issue of the journal Science Translational Medicine (paid subscription required).

The team led by Rady Children’s Institute president and CEO Stephen Kingsmore is seeking to improve the outcomes for babies born with genetic diseases, which the authors say are the leading cause of infant mortality in the U.S. and account for about 15 percent of cases in hospital intensive care units. Disease progression in infants is rapid, say the researchers, but current diagnostic techniques for genetic disorders are not always up to the task. Standard genome sequencing, for example, can take weeks to return a diagnosis, too slow for many seriously ill newborns.

A technique that can speed diagnostics of inherited diseases is rapid whole-genome sequencing designed to optimize and make the genetic analysis process more efficient. Kingsmore and colleagues in 2015 reported a 26-hour genetic diagnosis for infants in a research setting, an achievement that set a Guinness world’s record at the time for the fastest genetic diagnosis. But in hospital labs, it still takes 37 hours for genetic testing, plus 16 days for a complete diagnosis on average. In the new study, the Rady Children’s Institute team and colleagues from academic and company labs aimed to further tighten the process.

Among the practices employed in the accelerated process is clinical natural language processing, a branch of artificial intelligence that applies machine learning models to the analysis of medical records, particularly electronic health records. Medical records, even in automated systems, contain a good deal of free text in clinicians’ notes and images that are not as easily analyzed as structured data. Another machine-learning software package called Moon, made by the company Diploid in Leuven, Belgium, was used by the researchers to identify causal mutations in the medical records in about 5 minutes.

The team also employed algorithm-driven decision-support software that its developer, Fabric Genomics in Oakland, California says returns results in under an hour. In addition, the researchers used bead-based preparations of specimen samples, which standardizes identification of genetic transpositions, eliminating some of the normal sample preparation steps.

With artificial intelligence and other automated tools, the team’s accelerated and largely autonomous process completed genomic sequencing and phenotyping — identification of characteristics or traits associated with genetic factors — and subsequent interpretation of blood samples and records from 95 children with 97 genetic diseases in a median of 20 hours and 10 minutes, a time savings on average of nearly a day. Comparing the results of the diagnostics to expert reviews show a 97 percent sensitivity, or accurate diagnosis, and 99 percent precision.

The researchers also used their techniques with 7 seriously ill infants admitted to Rady’s intensive care unit. Of the 7 children, 4 could not be diagnosed either with autonomous or manual methods. The remaining 3 children, says the team, were all accurately diagnosed with the A.I.-based system, returning results that influenced their subsequent treatments.

“Using machine-learning platforms doesn’t replace human experts,” says Michelle Clark, statistical scientist and first author of the paper in a Rady Institute statement. “Instead it augments their capabilities. By informing timely targeted treatments, rapid genome sequencing can improve the outcomes of seriously ill children with genetic diseases.”

More from Science & Enterprise:

• Genetic Risk Analysis Shows Ethnic Bias
• Ancient DNA Studied for Mental Health Clues
• Crispr Diagnostics Company Launches with $35M Funding
• Genomics, A.I. Calculate Immune Health
• Population Data Show Genetics-Health Links

*     *     *