Researchers at The Hospital for Sick Children have developed a new machine learning model designed to improve the accuracy of prenatal genetic testing. Prenatal tests often identify genetic variants of uncertain significance, which creates stress and ambiguity for expecting parents. These variants are difficult to classify as harmful or harmless, leaving clinicians without a clear path for diagnosis.
The team’s work focuses on episignatures, which are specific chemical markers in DNA that signal the presence of genetic conditions. While scientists previously established many of these markers, they were historically tissue-specific, limiting their use in prenatal settings where only specific tissues like amniotic fluid are available.
By building a machine learning model capable of translating blood-derived episignatures into tissue-agnostic versions, researchers have opened a pathway for more precise diagnostics. In testing, the model successfully identified Down syndrome patterns across multiple tissue types. This suggests the technology can work effectively regardless of the sample source, potentially reducing the need for invasive testing procedures in the future.
This method marks a step forward in prenatal medicine by providing clearer information for families facing diagnostic uncertainty. The research highlights how computational tools can resolve complex genetic puzzles that currently exceed the manual diagnostic capacity of clinical genetics teams.