Machine learning advances accuracy in prenatal genetic testing

Machine learning is transforming prenatal genetic screening by improving the precision of diagnostic pipelines. Traditional methods, which rely on maternal age and ultrasound markers, often struggle with the high volume of noise in cell-free DNA (cfDNA) samples, particularly when the fetal fraction is low. New models, including Convolutional Neural Networks and Random Forests, are being used to better distinguish between fetal and maternal genetic signatures.

Researchers at The Hospital for Sick Children (SickKids) have developed a machine learning model designed to address variants of uncertain significance (VUS). Currently, clinicians are unable to classify approximately one-third of detected genetic changes as either harmful or benign. The new innovation utilizes episignatures—distinct chemical tags in DNA—to help determine if a variant is disease-causing. While previous platforms like EpigenCentral were limited to tissue-specific data, such as blood, these advancements aim to provide clarity across different samples, including amniotic fluid and placental tissue.