Fig. 2: Cross-validated cell line classification performance across passages.

a–c Classification performance using an equal number of cells from each passage for training and testing. d–f Classification performance using all cells from each passage for training and testing. For each cross-validation fold, balanced accuracies were calculated on test set cells from monoclonal wells, while impurity scores were calculated on test set cells from all wells. a, d Scatter plots illustrate the cross-validated classification performance of the three passages. Each data point represents the result of one cross-validation fold. Bar plots in (b, e) show the balanced accuracies (higher is better) of test cells from monoclonal wells, while (c, f) present the impurity scores (lower is better) calculated on test cells from all wells. Error bars represent the standard deviations. Cell line predictions for monoclonal wells and artificial pools in different passages are shown in (g–i), (j–l), and (m–o) for passage 0, passage 1, and passage 2, respectively. g, j, m Confusion matrices illustrating the cell line classifications of monoclonal wells. h, k, n Ratio of cells with different predicted cell line classes in different monoclonal wells and artificial pools. i, l, o Visualization of cell line predictions in randomly selected FOVs from monoclonal wells and artificial pools, where cells are color-coded by their predicted cell line classes.