samples. Details on how to do this can be found on the follow-
ing webpage: “tinyurl.com/KNIME-MS-ECS”.
- The downsampling factor that you use in the prediction phase
should correspond to the downsampling factor that was used
to train the pipeline. To train the pipeline we used a factor of
0.33 for data with pixel size 0.1072μm. If the pixel size in your
samples is different, you should scale the downsampling factor
for the prediction phase accordingly. For example, if your pixel
size is 1.5larger than ours, use a downsampling factor of 0.5.
- We have found that the standard settings of the PathFinder
plugin will give good results in many cases. If you observe
strong over- or under-segmentation you may change the Seg-
mentation parameters. Increasing the threshold for vertex,
membrane path, or membrane pixel detection will reduce
over-segmentation. Decreasing the small cell removal parame-
ter will also reduce over-segmentation.
- Image names do not need to match between folders, but make
sure that the order of images is identical between folders as the
PathFinder plugin will work its way down the file list.
- Cell segmentation will not be fully accurate. You can use the
Cell annotation tool (Fig.5) to correct small segmentation
errors that might have been made in the PathFinder tool. In
addition to cell segmentation correction, the “Cell annota-
tion” tool can be used to define different cell types. Options
are preset for early embryonic cell types, namely EVL (envel-
oping layer, #1), YSL (yolk syncytial layer, #2), DEL (deep
layer, #3), and Outside (#4) for regions outside of the embryo.
- Before opening the Cell annotation tool again to start with
your next image, you will need to Reset startup tools in Fiji
(press in the Fiji menu bar and then Restore Startup Tools).
- If you are not interested in assigning transcripts to single cells,
you do not need to use the membrane segmentation pipeline
to generate a cell mask. Instead, you can use the “Cell annota-
tion” plugin to outline your areas of interest and use the
resulting file as a mask or use no mask at all.
- To ensure reproducibility, we recommend to use the same
transcript detection thresholds for all images of the same sam-
ple (coverslip) that were acquired in the same imaging session.
- To ensure reproducibility, we recommend to use the same
parameters for nuclei segmentation and foci segmentation for
all images of similar embryonic stages that were acquired at the
same pixel size.
- Additional data files that are generated are: a file with the
analysis parameters (‘_parameters.txt’), a file with transcript
detection information per channel (‘spot*.txt’), a log file
smFISH and Automated Analysis in Zebrafish 161
