Non-invasively collected buccal cell mRNA and potential for a novel breast cancer signal.

Abstract

e12539 Background: Breast cancer remains the most frequently diagnosed cancer globally, despite current available diagnostic methods such as mammograms and ultrasounds. While DNA genetic testing is widely used clinically, messenger RNA (mRNA) analysis can provide complementary diagnostic information, particularly regarding inflammatory pathways associated with cancer development. Intercellular mRNA communication is an essential hallmark of chronic inflammation associated with cancer development, and it is important to create an integrative computational tool that will identify unmet personalized human needs related to development of chronic diseases. We present a novel screening procedure combining mRNA genetic markers with machine learning algorithms to enhance breast cancer detection. Methods: Buccal cheek swab samples were collected non-invasively from breast cancer patients and controls across 40 U.S. clinical centers, and a custom microarray chip was developed to analyze 48 mRNA-based cytokine biomarkers, selected for their association with inflammatory pathways and cancer development. Next-generation sequencing was performed using a Thermo Fisher Genestudio S-5 following cDNA target amplification. We used these data to train a linear support vector machine to classify subjects with respect to their cancer status. We were particularly interested in identifying a small subset of these biomarkers that could yield significant results to create a lower-dimensional, efficient classification approach. Results: Our analysis identified a collection of six-biomarker combinations yielding excellent results (F-score ≥0.85). We present the two top-performing biomarker combinations (see table), along with their respective performance metrics. Conclusions: Our results demonstrate the potential of mRNA biomarker panels, combined with machine learning, to generate a signal for detecting breast cancer from non-invasively-collected samples. The high precision and recall values suggest clinical utility for early detection. Summary of top-performing systems for detecting breast cancer. Systems used a combination of six biomarkers selected from a panel of 48 cytokines known to be associated with cancer or inflammatory pathways. BM1 BM2 BM3 BM4 BM5 BM6 P R F1 AUC SYSTEM 1 CASP9 IL1B IL6 RGS2 SLC40A1 STK11 0.892 0.913 0.9 0.897 SYSTEM 2 HAMP IL10 RGS2 SLC22A4 STK11 TGFB1 0.888 0.902 0.892 0.849