Fig. 3
Supervised data analysis of 131 mouse oocyte images with surrounded (SN) or not surrounded (NSN) chromatin organization. a The data analysis workflow first imports the data from the local directory where images are stored in respective subdirectories named SN and NSN. Vector-based embedding passes the data matrix to a cross-validation widget (Test and Score) that accepts a machine-learning method (logistic regression) as an additional input. The Test and Score widget displays the cross-validated accuracy (area under ROC curve—AUC, classification accuracy—CA, and harmonic average of the precision and recall—F1 score) (b) and sends the evaluation results to the Confusion Matrix widget (c). The Confusion Matrix widget provides information on misclassification. In this example, 65 of the 69 SN oocytes were classified correctly. Selection of this particular cell in the Confusion Matrix triggers sending these images and their descriptors further down the workflow to an Image Viewer (d) and, as a subset of data points, to the MDS widget that performs multi-dimensional scaling (e). Just like in Fig. 1, the MDS widget shows a planar projection of data points (images) and highlights, in this case, the image points selected in the Confusion Matrix. Altogether, the components of this workflow are used to quantitatively evaluate the expected performance of machine-learning models through cross-validation and to support further exploration of correctly and incorrectly classified images
