Machine learning algorithms enhance the specificity of cancer biomarker detection using SERS-based immunoassays in microfluidic chips

Abstract

Specificity is a challenge in liquid biopsy and early diagnosis of various diseases. There are only a few biomarkers that have been approved for use in cancer diagnostics; however, these biomarkers suffer from a lack of high specificity. Moreover, determining the exact type of disorder for patients with positive liquid biopsy tests is difficult, especially when the aberrant expression of one single biomarker can be found in various other disorders. In this study, a SERS-based protein biomarker detection platform in a microfluidic chip and two machine learning algorithms (K-nearest neighbor and classification tree) are used to improve the reproducibility and specificity of the SERS-based liquid biopsy assay. Applying machine learning algorithms to the analysis of the expression level data of 5 protein biomarkers (CA19-9, HE4, MUC4, MMP7, and mesothelin) in pancreatic cancer patients, ovarian cancer patients, pancreatitis patients, and healthy individuals improves the chance of recognition for one specific disorder among the aforementioned diseases with overlapping protein biomarker changes. Our results demonstrate a convenient but highly specific approach for cancer diagnostics using serum samples.