Fig. 4

Identification of low quality cells. a Visualizing low and high quality cells with traditional and feature-based PCA method. The feature-based method makes it easier to detect low quality cells visually as most of them are outliers. b Accuracy measurements to evaluate the performance of each method. Sensitivity is defined as the proportion of correctly identified low quality cells. Specificity is defined as the proportion of correctly identified high quality cells. SVM outperforms all other methods as it has reasonable sensitivity and high specificity. c Comparing the effect of all versus common features upon the trained on SVM: all features result in higher sensitivity and specificity. F-score is defined as the harmonic mean between sensitivity and specificity. d Linear SVM feature weights illustrated as word clouds. Red features are informative for low quality and green features for high quality cells