Researchers at Minnesota-based Astrin Biosciences, working with collaborators at the University of Minnesota Masonic Cancer Center, Mayo Clinic, and Allina Health, reported that a machine learning classifier applied to deep plasma proteomics distinguished breast cancer from healthy controls with 92.6% sensitivity and 92.3% specificity in an independent validation cohort, according to a study published July 30 in Breast Cancer: Targets and Therapy. The test is being developed as a supplemental screening tool for women with dense breast tissue, where mammography sensitivity falls to 47–69%.
While mammography remains the standard screening method, its performance declines in women with heterogeneously or extremely dense breasts. MRI, contrast-enhanced mammography, and automated breast ultrasound improve detection but are limited by cost, scalability, and higher false-positive rates. Circulating tumor DNA-based liquid biopsies, including Grail's Galleri test, have shown lower sensitivity for early-stage breast cancer because of limited tumor DNA shedding, while conventional protein biomarkers lack sufficient screening accuracy.
The retrospective study analyzed plasma from 1,242 women, including 661 healthy controls and 581 untreated breast cancer patients spanning Stages 0–IV. After quality control, the dataset was divided into 845 training samples and 397 independent validation samples. Using automated sample preparation, deep mass spectrometry profiling, and a fairness-constrained machine learning classifier trained on approximately 8,400 proteomic features, the model exceeded its pre-specified performance targets, achieving 92.6% sensitivity and 92.3% specificity in the validation cohort. The authors also reported 100% detection of invasive lobular carcinoma cases in the validation set.
Gene set enrichment analysis identified enrichment of epithelial-to-mesenchymal transition, PI3K-AKT, KRAS, and WNT/β-catenin signaling pathways, while inflammatory pathways were largely absent, suggesting the classifier primarily detects tumor-derived biology rather than nonspecific immune activation.
A Monte Carlo simulation comparing supplemental screening strategies for women with dense breasts suggested that adding the Astrin test to mammography could detect 93% of cancers missed by mammography while generating more than 10-fold fewer false positives than MRI or contrast-enhanced mammography alone. The authors emphasized that these findings are based on modeling and require prospective clinical validation.