Machine learning reveals mesenchymal breast carcinoma cell adaptation in response to matrix stiffness
Fig 4
Identification of cell morphs using machine learning.
(a) The results of HC applied to 826 cells based on similarities between single-cell profiles consisting of 150 morphological and contextual features. Left column indicates three cell morphs (green, blue, and red) identified in the total population of cells. Each row of the heatmap corresponds to a single cell and each column represents a single feature. Subgroups of features are depicted on the bottom. (b) Representative objects (medoids) of each of the three identified cell groups. (c) Top six features ranked by their contribution towards cluster separation. Distributions of the values are reported using box plots: a box shows the median value, first and third quartiles, whiskers indicate median +/-1.5 * IQR. Significance of the difference assessed by Welch’s t-test (‘***’: p < 0.001). (d) Projection of the data on the first two PCs and visualisation of the identified cell morphs. “×” markers indicate the centroids of the groups. (e) Proportion of cells of each morph across stiffness values. Bar plots represent average proportions, error bars indicate 95% confidence intervals, values calculated by grouping cells by images. The inset shows the total number of cells in each cluster.