With Machine Learning, Researchers Improve Potential Blood-Based Breast Cancer Test

By: Celeste L. Dixon
Posted: Monday, August 19, 2024

Using machine learning, investigators have identified 51 cytosine-phosphate-guanine (CpG) sites—from a possible 649,688—that together constitute a blood-based DNA methylation profile. They believe this profile may effectively distinguish patients with breast cancer from healthy controls better than any of the four previously reported such profiles. This one achieved an area under the curve score of 0.823 on an independent test set, reported Ann S.G. Lee, DPhil, of National Cancer Centre Singapore, and colleagues, in the journal Clinical Epigenetics.

Finding precise biomarkers for early noninvasive breast cancer detection has been an important unmet need, wrote the team. Developing an accurate blood-based biomarker assay may potentially reduce overall breast cancer mortality, but it could also considerably reduce the costs associated with false-positives and overdiagnosis in current screening programs, they noted.

This study included DNA methylation profiling performed on a cohort of 524 Asian Chinese individuals: 256 patients with breast cancer and 268 age-matched healthy controls. The authors used the Infinium MethylationEPIC array.

Previously, the search for breast cancer–associated methylation profiles in peripheral blood DNA has been targeted mostly at cancer genes, noted Dr. Lee and co-researchers. However, “our results suggest that the inclusion of DNA methylation of immune-related genes or pathways could improve the performance of peripheral blood screening for breast cancer,” they stated. “Enrichment analyses of transcription factor–DNA binding and functional pathways of genes associated with the 51 CpGs both suggest the host immune response against cancer may play a role in driving the difference between methylation profiles of breast cancer cases and healthy controls.”

The team also pointed out that the best-performing previously reported breast cancer–associated methylation profile had been trained on a predominantly European population, yet it performed “reasonably well” in this study’s Asian patient cohort.

Disclosure: The study authors reported no conflicts of interest.

Clinical Epigenetics

---