The European Molecular Biology Laboratory (EMBL) and the Technical University of Munich have created an algorithm that uses deep learning to predict how rare genetic variants affect organisms. Their research, “Integration of Variant annotations using deep set networks boosts rare variant testing,” has appeared in Nature Medicine.
Each person’s genetic makeup differs from others by millions of individual building blocks. We call these differences in the genome variants. Many of these variants have a connection to specific biological traits and diseases. Scientists figure out these connections using what we call genome-wide association studies.
“Uncommon variants often have a much bigger impact on how a biological trait or a disease shows up,” says Brian Clarke, one of the lead authors of this study.
To predict the impact of uncommon genetic variations more, research groups led by Oliver Stegle and Brian Clarke at DKFZ and EMBL, along with Julien Gagneur at the Technical University of Munich, have created a risk assessment tool that uses machine learning. The researchers named their method “DeepRVAT” (rare variant association testing). This approach stands out as the first to apply artificial intelligence (AI) in genomic association studies to unravel rare genetic variants.
The researchers tested DeepRVAT on genome data from the UK Biobank. The testing method identified 352 associations with genes involved for 34 tested traits, which are disease-relevant blood test results. This performance far surpassed all existing models. The outcomes from DeepRVAT showed strong reliability and better reproducibility in independent data compared to alternative methods.
DeepRVAT also has an important use to assess genetic predisposition to certain diseases. The researchers combined DeepRVAT with polygenic risk scoring based on more common genetic variants. This combination enhanced the accuracy of the predictions for high-risk variants.
Moreover, DeepRVAT identified genetic links for many illnesses—including several heart diseases, cancer types, metabolic and brain disorders—that previous test had missed.
“DeepRVAT could push personalised medicine forward. Our approach works for any kind of trait and can easily team up with other testing methods,” says Oliver Stegle, a physicist and data scientist. His group now aims to further evaluate the risk assessment tool in big trials as soon as possible and put it to use.
The scientists are already talking with the INFORM organizers. This study aims to use genetic data to find custom treatments for kids with cancer who get sick again. DeepRVAT could help to uncover the genetic reasons for some childhood cancers.
Reference:
Cancer G. How do rare genetic variants affect health? AI provides more accurate predictions.
DOI: 10.1038/s41588-024-01919-z. www.nature.com/articles/s41588-024-01919-z
