Journal: PLOS ONE

Article Title: Machine learning evaluation for identification of M-proteins in human serum

doi: 10.1371/journal.pone.0299600

Figure Lengend Snippet: (A) , Capillary zone electrophoresis (CZE) from a sample containing M-protein. The x-axis is divided into 300 time points (features), representing the migration time during SPEP. The most important features for detecting M-protein based on the SHAP calculations are highlighted in yellow, clearly verifying the gamma globulin fraction and part of the beta-2 globulin fraction as the most important for detecting M-protein. (B) , The ten most important features (from a total of 300) for detection of M-protein. Each feature value corresponds to the migration time on the x-axis of the CZE. The x-axis shows the mean SHAP value. (C) , SHAP values for a sample with no M-protein (top) and with M-protein (bottom). Red numbers indicate positively correlated features for detecting M-protein and blue numbers indicate negatively correlated features. Probability scores for the presence of M-protein in the samples are shown in bold numbers. Features with their corresponding SHAP values are indicated below each panel.

Article Snippet: All samples were analyzed on both an automated capillary electrophoresis system (Capillarys 2 Flex Piercing, SEBIA, Issy-les-Moulineaux, France), for quantification, as well as by a semi-automated agarose gel technology system (serum gel electrophoresis) (Hydrasys 2 System and PENTA PN 1260 kit, SEBIA, Issy-les-Moulineaux, France) [ , ] to not miss small fractions of M-protein.

Techniques: Electrophoresis, Migration