Currently, many patients receive inconclusive results from genetic testing, leaving them without a clear diagnosis or a path forward for treatment, but a new model developed by researchers at Duke University could help increase the number of patients receiving positive test results, providing them with valuable information that could guide treatment decisions. Results were published in Circulation: Genomic and Precision Medicine.
By incorporating factors such as minor allele frequencies, gene expression in cardiovascular tissues, splicing effects, conservation scores, and the location of the variant relative to known pathogenic variants, their model predicts with high accuracy whether a variant is likely to be pathogenic.
The team used the model on the CathGen cohort – a collection of patients undergoing cardiac catheterization – and identified individuals carrying likely pathogenic variants. “A majority of those people carrying predicted pathogenic variants,” Ramaker said, “displayed symptoms that would allow us to provide a diagnosis based on genetics.” It will also help clinicians prescribe medicines to treat individual conditions more effectively, as some treatments are gene specific.
While the model was developed with a focus on cardiovascular tissue, it can be adapted for other diseases. "You could input a tissue relevant to your disease of interest, like cancer, and modify the model accordingly," Ramaker said. This opens up the possibility for researchers in other fields to fine-tune the model for their own needs.
Next, the team plans to work on expanding the use of this tool to larger genetic biobanks, such as the UK Biobank and All of Us, and OneDukeGen. They are also applying it to investigate genetic variants linked to cardiac amyloidosis, a rare inherited disease that leads to plaque buildup in the heart, similar to how Alzheimer's disease affects the brain.
”There are a lot of machine learning tools to predict variant pathogenicity,” Ramaker said, “but this is the first one that is tissue specific and publicly available.”