Machine Learning Approach Expands Epigenetic Testing for Prenatal Genetic Disorders

This uncertainty can leave expectant families without clear guidance, with about one-third of detected genetic changes lacking a definitive interpretation. Clinical laboratories also face tissue-related limitations because many epigenetic signatures have been developed using blood samples. New findings demonstrate a machine learning approach designed to extend epigenetic variant assessment across both prenatal and postnatal tissues.

Researchers at The Hospital for Sick Children (SickKids; Toronto, ON, Canada) have developed a tissue-agnostic machine learning model that classifies uncertain genetic variants using epigenetic “episignatures.” The team’s work builds on EpigenCentral, a platform launched in 2020 that uses blood-derived episignature data to help clinicians determine whether a variant is disease-causing or benign. While the lab has helped establish more than 60 episignatures—most clinically proven for diagnosis—these patterns were historically tissue-specific, limiting their use when only amniotic fluid or placental samples were available in prenatal settings.

The new method uses machine learning to transform blood-derived episignatures into “tissue-agnostic” profiles that remain informative regardless of sample origin, including prenatal specimens. Episignatures, which reflect DNA methylation patterns linked to specific genetic conditions, can therefore be applied across multiple tissue types. By addressing tissue specificity, the approach is designed to support variant interpretation when traditional sample types are not accessible.

To demonstrate proof of concept, the team generated a blood-derived episignature using samples from 266 individuals with Down syndrome. They then trained the model on publicly available DNA methylation data from 850 individuals with and without Down syndrome, spanning six prenatal and postnatal tissue types. The model recognized the Down syndrome pattern in every tissue type tested, indicating that a blood-derived episignature can be converted into one that detects a disease-specific signal across diverse tissues.

Findings of the study were published in The American Journal of Human Genetics. The researchers note that the framework could support testing with a broader range of specimens, such as saliva and oral swabs, potentially reducing the need for blood or other tissue. The work aligns with Precision Child Health at SickKids, which emphasizes individualized diagnosis and prediction for children with rare disorders.

“We’re thrilled that our model can bring a new level of precision to prenatal testing, where so many questions remain to be answered. This machine learning approach is also a building block to study many disorders where tissues are inaccessible, which would support rapid translation into the clinic,” said Dr. Rosanna Weksberg, Clinical Geneticist and Senior Associate Scientist, Genetics & Genome Biology at The Hospital for Sick Children (SickKids).

“When families are told of a potential neurodevelopmental disorder, our goal with this approach is to provide more information, as detailed as possible, to support them. We want to reduce the diagnostic odyssey that many of these patients go through for rare disorders,” said Dr. Sanaa Choufani, Senior Research Associate.

Related Links
The Hospital for Sick Children