Supplementary Figure 5: Performance of 3D fluorescence image prediction from transmitted-light microscopy (bright-field) image inputs of different cell types.

a, Representative DNA predictions on 3D bright-field input images of, from the top, HEK-293 cells, cardiomyocytes, and HT-1080 cells using a DNA model trained on hiPSC images (model predicting DNA with an extended training procedure, “DNA+”, see Methods) and a DNA model trained on HEK-293 images ("m-HEK-293"). From left to right, columns are the bright-field input image, the target fluorescence image, the predicted fluorescence image using the DNA+ model, and the predicted fluorescence images using the m-HEK-293 model. The center z-, y-, and x-slices are shown for each 3D image (denoted by the cyan, green, and red lines, respectively, in the leftmost image). Scale bar is 20 µm. b, Distributions of the image-wise Pearson correlation coefficient (r) between target and predicted test images of different cell types using DNA+ and m-HEK-293 models. Each target/predicted image pair in the test set is a point in the resultant r distribution; the 25th, 50th, and 75th percentile image pairs are spanned by the box for each indicated structure, with whiskers indicating the last data points within the ×1.5 interquartile range of the lower and upper quartiles. Outliers for each distribution are shown as circles. Box colors for each input images cell type indicate which model was applied, DNA+ or m-HEK-293. The numbers of images (n) in the hiPSC, HEK-293, cardiomyocyte, and HT-1080 distributions are 20, 18, 8, and 10, respectively. For more details see Methods.