Deep Proteomic, AI Breast Cancer Detection Test Demonstrates High Performance

A blood-based early breast cancer detection test demonstrated high performance across different cancer stages and molecular and pathological subtypes.^1,2

The test, developed by Astrin Biosciences, utilizes deep proteomic profiling and a machine learning classifier to identify cancer-specific protein signatures in plasma. The central challenge addressed is the significant limitation of mammography, which has a sensitivity as low as 30% in the 50% of women with dense breast tissue.

Results using Certitude were announced at the 2025 San Antonio Breast Cancer Symposium (SABCS) on December 11, 2025.

The assay demonstrated high performance in both training and a blinded validation set, showing robust sensitivity and specificity across key subgroups. The model achieved an area under the curve of 0.961. The classifier achieved 92.3% sensitivity at 92.6% specificity.

Sensitivity remained high and above 84% across all cancer stages. The test maintained high sensitivity across all pathological and molecular subtypes. For triple-negative breast cancer, sensitivity was greater than 90%. The study found no demographic confounders, and analysis showed consistent sensitivity and specificity across different age groups and specimen sources.

Gene set enrichment analysis was conducted on the breast cancer samples. This analysis confirmed that the assay was successfully identifying cancer-associated proteins and pathways, including enrichment for epithelial-to-mesenchymal transition and PI3K-AKT signaling pathways.

“These results reinforce the promise of combining deep proteomics with [artificial intelligence] to transform how breast cancer is detected,” said Justin M. Drake, PhD, lead author or the study and chief science officer at Astrin Biosciences in a news release. “By identifying disease signals at the earliest stages, we can help more women benefit from timely intervention and improved outcomes.” 

Study Cohort and Demographics

The classifier was developed and validated using banked plasma samples from treatment-naive patients with breast cancer and demographically matched healthy donors. Samples were procured from 2 different vendors, ProteoGenex and BioIVT.

In the training set, there were 466 healthy patients and 379 patients with breast cancer. In the validation set, there were 195 healthy patients and 202 patients with breast cancer. The mean age of healthy patients in the training set was 55.4 years vs 57.2 years in patients with cancer. In the validation set, the mean age of healthy patients was 56.4 years vs 59.8 years in patients with cancer. In both sets, the majority of patients were White. In both sets, the majority of patients with cancer had stage 2 (40.1%).

Assay Overview and Methodology

The assay analyzes plasma to identify distinctive, cancer-specific protein signatures. Using a superparamagnetic bead solution, proteins are isolated from plasma samples and are subsequently digested to create peptides. A machine learning classifier was developed using data from 845 women. The model employs an exponentiated gradient method with L2-norm regularized logistic regression.

To ensure robustness and mimic real-world conditions, samples were randomly assigned to either a training or validation set. The validation samples were blinded during sample preparation and analysis and were batched separately from the training samples.

Clinical Need and Existing Diagnostic Limitations

While mammography has been instrumental in enabling early diagnosis for many, its efficacy is severely compromised in women with dense breast tissue. This demographic, which constitutes half of all women, experiences mammography sensitivity as low as 30%.

Other liquid-based biopsy techniques are in development but are not yet ready for clinical use. Specifically, methods based on nucleotide assessment from plasma have shown mixed results, with a reported 87% sensitivity for late-stage (III–IV) breast cancer but only 20% sensitivity for early-stage (I–II) disease.

The FDA's 2024 mammography regulation has increased patient awareness of the challenges associated with dense breast tissue. However, this has also led to heightened patient anxiety, as it highlights a significant problem without offering an immediate, accessible solution.³

REFERENCES

1.Horrmann A, Travadi Y, Carey J, et al. Deep proteomics and AI classifier for early breast cancer detection. Presented at: San Antonio Breast Cancer Symposium; December 11, 2025; San Antonio, TX.Abstract: 1274.

2.Astrin Biosciences launches Certitude Breast, the first non-imaging breast cancer detection test. News release. Astrin Biosciences. December 3, 2025. Accessed December 15, 2025. https://tinyurl.com/59mydyz6

3.Mammography Quality Standards Act (MQSA) and MQSA program. US FDA. Accessed December 15, 2025. https://tinyurl.com/yhp8ey5n