| Shallow classification using manually defined features.

(A) Image overview of 1321N1 cells after CP staining acquired at higher magnification (Plan Apo VC 60×A WI DIC N2, NA = 1,20). Scale bar is 30µm. (B) UMAP dimensionality reduction, color coded for either cell type condition (left) or replicate (right). (C) Feature importance deducted from the UMAP (feature maps) and RF (Gini importance). Three exemplary feature maps are highlighted alongside the quantification per cell type. (D) Random Forest classification performance with confusion matrix and Gini importance. All features used in the UMAP are used for RF building.

| Convolutional neural network classification of monoculture conditions.

(A) Schematic of image pre-processing for CNN classification. (B) CNN accuracy and confusion matrix on cell type classification in monocultures. (C) Shallow vs. deep learning accuracy for varying training set sizes. Dots represent different training iterations (cross-validation). Train-validation-test dataset selection was performed with the same random seed at 60-10-30 ratios. (D) The impact of variability in the training dataset on model accuracy for shallow vs. deep learning. Mann-Whitney-U-test, p-values resp. 7,47e-6 and 1,96e-4. (E) The performance of deep learning classifiers trained on single replicates (low variability) and mixed replicates (high variability) on unseen images (independent replicates). Kruskal-Wallis test on single training condition, p-value of 0,026 with post-hoc Tukey test. Mann-Whitney-U-test on mixed training, p-value of 7,47e-6. (F) GradCAM map highlighting the regions on which the CNN relies for making its prediction. (G) Performance accuracy of the CNN when only a selection of image channels is given as training input.

| The impact of regional information on CNN performance in relation to culture density.

(A) Definition of different culture density categories with corresponding images. (B) The effect of culture density (number of cells within the FOV) on prediction accuracy, for each of the different input regions. Kruskal-Wallis test for each of the regions, statistically significant decrease (p-value 2,07e-5) between the most dense condition with all other density categories for the cell region. (C) CNN accuracy for different input regions. Kruskal-Wallis test, p-value of 4,29e-4 with post-hoc Tukey test. (D) GradCAM intensity plots highlighting the regions on which the CNN relies for making its prediction in nucleocentric (top) and nuclear (bottom) crops.

| Cell type prediction in mixed cultures by post-hoc ground truth identification.

(A) Schematic overview of virtual vs. physically mixed cultures and the subsequent CNN models. For virtual mixing, individual cell types arise from distinct cultures, the cell type label was assigned based on the culture condition. For physical mixing, the true phenotype of each crop was determined by intensity thresholding after post-hoc staining. Three model types are defined: Mono2Mono (culture-based label), Co2Co (cell-based label) and Mono2Co (trained with culture-based label and tested on cell-based label). (B) Determining ground-truth phenotype by intensity thresholding on IF (top) or pre-label (bottom). The threshold was determined on the intensity of monocultures. Only the pre-labelled ground-truth resulted in good cell type separation by thresholding for 1321N1 vs. SH-SY5Y cells. (C) Mono2Mono (culture-based ground truth) vs. Co2Co (cell-based ground truth) models for cell type classification. Mann-Whitney-U-test p-value 0,0027. (D) Monoculture-trained models were tested on mixed cultures. Pretrained models were trained on independent biological replicates. These could be finetuned by additional training on monoculture images from the same replicate as the coculture. This was shown to reduce the variation between model iterations (Median performance: Mann-Whitney-U-test, p-value 0,0015; Coefficient of variation: Mann-Whitney-U-test, p-value 3,48e-4)

| iPSC-derived differentiation staging using morphology profiling.

(A) Schematic overview of guided vs. spontaneous neural differentiation. (B) Representative images of morphological staining and post-hoc IF of primed and differentiated iPSC-derived cultures (guided differentiation). Ground-truth determination is performed using TUBB3 (for mature neurons) and Ki67 (mitotic progenitors). Scale bar is 200µm. (C) Fraction of neurons vs. NPC cells in the primed vs. differentiated condition as determined by IF staining. Upon guided differentiation, the fraction of neurons increased. (D) Following feature extraction (handcrafted features), supervised UMAP dimension reduction shows separation between the primed vs. differentiated cultures. (E) CNN performance when classifying neurons (Ki67-/TUBB3+) vs. NPC (Ki67+/TUBB3-) cells using either a condition-based or cell-type based ground truth. Mann-Whitney-U-test, p-value 4,04e-4. (F) Fractional abundance of predicted cell phenotypes (NPC vs. neurons) in primed vs. differentiated culture conditions using the cell-based CNN. (G) Representative images of spontaneously differentiating neural cultures. Scale bar is 200µm. (H) Handcrafted features were extracted from the morphological images. These morphological profiles were plotted onto the UMAP feature space created using the guided differentiation (panel D). Plotting the points of spontaneously differentiating cultures onto this space, revealed gradual displacement of the morphological profile from the NPC towards the neuronal cluster with increasing differentiation time. (I) Quantification of the number of points in each cluster of panel H. (J) Prediction of differentiation status using the cell-based CNN model trained on guided differentiated culture.

| iPSC cell type identification using morphology profiling.

(A) Representative images of iPSC-derived neurons, astrocytes and microglia with morphological staining. Scale bar is 200µm. (B) Prediction accuracy of a CNN trained to classify monocultures of iPSC-derived astrocytes, microglia and neurons (panel A) with confusion matrix (average of all models). (C) Representative images of a mixed culture of iPSC-derived neurons and microglia. Ground-truth identification was performed using IF. (D) CNN accuracy of a model trained to identify both cell types in mixed culture mounted up to 98%.

Specifications of morphological staining composition.

Used antibodies.

Specifications of the used laser lines, excitation and emission filters.

Python packages used for image and data analysis alongside the software version.

| The influence of cell density on segmentation quantity and quality.

(A) Representative images of 1321N1 cells with increasing density alongside their cell and nuclear mask produced using resp. Cellpose and Stardist. (B) the number of ROIs detected in comparison to the ground truth (manual segmentation) (determined as intersection over union (IoU) < 0,15). (C) ROI detection quality (IoU) for increasing cell density for cell and nuclear masks.

| Fluorescence quenching over time using LiBH4.

(A) Images before and after quenching for all 4 fluorescence channels. (B) Time curve of normalized image-level fluorescence intensity during incubation with 1mg/ml LiBH4.

| Methods.

(A) Pipeline used for morphological profiling in mixed cultures. (B) Overview of the steps within the image analysis pipeline. (C) Evaluation of accuracy, true negative and true positive rate during CNN training (1321N1 vs. SH-SY5Y in monocultures) across all 50 epochs.

| Additional figures.

(A) PCA dimension reduction on handcrafted features extracted from monocultures of 1321N1 and SH-SY5Y cells. In contrast to UMAP dimension reduction, this linear approach does not result in clear cell type clustering. (B) Examples of misclassified ROIs. (C) Definition of cell regions given as training input for nuclear and nucleocentric model training. (D) Decreasing nuclear area in function of culture density. (E) Evaluation of cell phenotype prediction on an image of differentiating neurons (guided differentiation). The predicted phenotype is shown by the color of the square around each ROI. The distinctive phenotype is shown in the insets. (F) Additional GradCAM images for both cell type classes. (G) Predicted neuron-to-NPC ratio in spontaneously differentiating cultures as predicted by the condition-based classifier (Fig. 5E) in contrast to the cell-based classifier performance on the same dataset (Fig. 5H).