A Blood Test for Early Breast Cancer Detection Using Single Molecule Arrays and Machine Learning
MetadataShow full item record
CitationHsin, Tien-yu. 2019. A Blood Test for Early Breast Cancer Detection Using Single Molecule Arrays and Machine Learning. Master's thesis, Harvard Medical School.
AbstractEarly breast cancer diagnosis is challenging due to the high false-positive rate and a lack of molecular classification of the current, imaging-based mammography. An estimated 50% of women screened annually in the U.S. experience a false-positive result, and of these women, 7%- 17% will receive unnecessary biopsies. Since tumors, even at early stages, secrete molecules into the bloodstream, a liquid biopsy has the potential to provide an alternative classification method for early detection. Concurrently, liquid biopsies could provide molecular information which would facilitate subtyping and help guide treatment. Although biomarkers can be overexpressed in breast cancer tissues, many clinically-relevant proteins are present at sub pg/mL concentrations in the blood and are therefore difficult to measure. To overcome this challenge, the Walt lab has previously developed a Single Molecule Array (Simoa) platform to achieve protein concentrations detection at a 10-19 M level compared to the conventional ELISA’s ability of 10-12 M detection. By labeling and isolating each immunocomplex into femtoliter-sized wells, we can digitally count the number of molecules in a sample. This assay format has been automated to reduce human error and increase accuracy. Additionally, due to the high sensitivity, only a low sample volume of 25 microliter or less is required. We collected 120 blood samples from healthy controls and 102 blood samples from untreated breast cancer patients. Testing a panel of 10 biomarkers on these blood samples, we achieve an AUC of 93% for breast cancer detection, which significantly outperforms mammograms. Whether Simoa confers robust prediction for subtypes and stages remains to be validated.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:42057401