MSACL 2024 Abstract

Introduction: Breast cancer (BC) is the most prevalent cancer among women worldwide and ranks as the second leading cause of cancer-related deaths. Fortunately, the prognosis for most breast cancers is very good when detected early and therefore screening method for timely detection of BC is vital for improving survival rates. The hope is that early screening of BC can be accomplished using non-invasive techniques such as analyzing blood samples with a targeted proteomic assay.

Objectives: The objective of this study was to determine protein biomarkers with diagnostic and prognostic value in human plasma to improve early BC diagnosis.

Methods: The absolute concentration of 131 proteins was determined in 291 biobank-based plasma samples, from BC patients and healthy controls. The panel of proteins were quantified using PeptiQuantTM protein human plasma kits containing paired labeled and unlabeled peptide standards. The plasma samples were proteolytically cleaved and concentrated by solid-phase extraction using a liquid handling robot and analyzed with ultra performance liquid chromatography-multiple reaction monitoring-mass spectrometry (UPLC-MRM-MS) assay. Data analyses were conducted with MassLynx and Skyline,while SIMCA Pro-17, R studio and Python were used for multivariate data analyzing and machine learning. A predictive machine learning models, including, random forest (RF) and support vector machine (SVM), were developed using the SciKit-Learn library in Python, to distinguish between BC samples and healthy controls while also considering various subgroups and clinical information. This study applied an ensemble classifier with soft voting to the dataset. The performance of the classifiers was evaluated using confusion matrices including accuracy, precision, recall, F1-score, and ROC curves.

Results: The targeted proteomics assay was successfully implemented for the absolute quantification of 131 proteins in human plasma. A total of 106 proteins were quantifiable in the biobank-based plasma samples with acceptable precision and accuracy and above the lower limit of quantification. The plasma samples were analyzed in separate batches, each containing 25 cases and 25 controls with a negligible batch difference. The dataset containing measurements of the 106 proteins from the 291 biobank-based plasma samples was divided into 233 (80%) samples for training purposes and 58 samples (20%) for testing. Cross-validation involved splitting the dataset into multiple subsets, training the models on different combinations of these subsets, and evaluating their performance. The predictive classifier models, RF and SVM, classified cases with 60% prediction accuracy and AUCs between 0.71 and 0.75. The soft voting ensemble classifier combining RF and SVM achieved 70% prediction accuracy and an Area Under Curve (AUC) score of 0.75. These results indicate that the models have a reasonable discriminatory power to classify cases from controls. The results also highlight promising prospects for early breast cancer diagnosis, especially in the comparison between sporadic BC and BC from BRCA2 mutation carriers, where the ensemble classifier achieved an AUC of 0.81.

Conclusions: Preliminary results indicate a potential difference in protein composition between plasma samples from BC patients and healthy controls that might suggest candidate biomarkers for early BC diagnosis.