There is an expectation that each spot_id in an IntensityTable is a unique identifier for each feature. There is a bug when using build_traces_sequential, which is used by SimpleLookupDecoder and possibly PerRoundMaxChannel when running a FindSpotsAlgorithm (e.g. BlobDetector.run()) with no reference_image. In this case, spots are found in each image volume. So if there are 3 rounds and 3 channels in the experiment, spots are found in 9 image volumes. And for each image volume the spot_ids are indexed starting from 0.
With build_traces_sequential, the PerImageSliceResults are simply concatenated with no reassignment of spot_ids, leading to spots from different image volumes sharing the same spot_id.
This impacts downstream steps in the pipeline like AssignSpots.label() , which uses the spot_id to label spots with cell_id. What is happening there is AssignSpots.label() is getting the spot_id of spots that are in each cell, and then labeling the spots using their row_index aka features.
mattcai
commented
4 years ago
Description
There is an expectation that each spot_id in an
IntensityTableis a unique identifier for each feature. There is a bug when usingbuild_traces_sequential, which is used bySimpleLookupDecoderand possiblyPerRoundMaxChannelwhen running aFindSpotsAlgorithm(e.g.BlobDetector.run()) with noreference_image. In this case, spots are found in each image volume. So if there are 3 rounds and 3 channels in the experiment, spots are found in 9 image volumes. And for each image volume the spot_ids are indexed starting from 0.With
build_traces_sequential, thePerImageSliceResultsare simply concatenated with no reassignment of spot_ids, leading to spots from different image volumes sharing the same spot_id.This impacts downstream steps in the pipeline like
AssignSpots.label(), which uses the spot_id to label spots with cell_id. What is happening there isAssignSpots.label()is getting thespot_idof spots that are in each cell, and then labeling the spots using their row_index akafeatures.Steps/Code to Reproduce
code to reproduce bug
Expected Results
decoded_intensitiesshould match featuresActual Results