Key points:
- People with cancer have different amounts of repetitive DNA—called Alu elements—detectable in their blood plasma relative to people without cancer.
- Researchers combined a blood liquid biopsy test that can detect chromosome copy number alterations with an unbiased machine learning approach called A-PLUS (Alu Profile Learning Using Sequencing) to accurately distinguish cancer from noncancer.
- Their technique reached 98.9% specificity and detected 41% of cancer cases missed by eight existing biomarkers and previous aneuploidy testing.
People with cancer have different amounts of repetitive DNA than people without cancer. A new machine learning technique, detailed in Science Translational Medicine, can measure repetitive DNA from a blood draw and has the potential to detect cancer early.
The changes in repetitive DNA—called Alu elements—occur in people’s blood plasma regardless of where their cancer originated. Researchers previously developed a blood liquid biopsy test that can detect aneuploidy—chromosome copy number alterations—in cancer. They then combined this test with an unbiased machine learning approach called A-PLUS (Alu Profile Learning Using Sequencing) to accurately distinguish cancer from noncancer.
The team collected samples from 3,105 people with 11 different types of solid cancers and 2,073 people without cancer. When testing samples and any replicates, the researchers found that the model reached 98.9% specificity and greatly reduced the number of false-positive results.
In a separate validation cohort, researchers determined that the addition of Alu elements to the machine learning model helped to detect 41% of cancer cases missed by eight existing biomarkers and previous aneuploidy testing.
“This research shows that counting repetitive lengths of DNA in blood plasma is cost-effective and enhances early cancer detection,” explained lead author Christopher Douville, professor at Johns Hopkins University.
Douville and his team imagine their Alu-based cancer detection as a complement to the other tests available to clinicians. To bring this closer to reality, they will prioritize which biomarkers seem the most promising and aggregate them together.