Mapping algorithm failed because of application error

  • Operating system and web browser: Windows 11 Pro
  • MapMyCells run ID: 1701350936199-4a85a24a-e962-4c15-903b-a3f9fb0e4d3e
  • Information on the data: Mouse fluorescent in-situ data converted into a count matrix, followed by a Seurat object in R. Column names consist of 10 genes denoted with Ensembl IDs and rownames are cell IDs. Data consist of puncta counts for each cell per gene (e.g. Row contains information for cell
    C B1 RZI MONTAGE P13190. The cell contains 158 transcripts of ENSMUSG00000028222, 3 transcripts of ENSMUSG00000003657, and so on for all 10 genes). There are no negative numbers in the dataset. The AnnData object contains n_obs × n_vars = 10878 × 10 with a total size of 1MB.
  • Instructions on how to reproduce the issue: Uploaded the h5ad file to the website. Mapping is submitted, input file is validated, algorithm is run, then fails after about 3-5 mins.

Thank you for your help!

Hi @pwashin6,

The executive summary:
Your dataset does not contain all of the marker genes our mouse taxonomy expects. This causes hierarchical mapping to fail. Correlation mapping should still pass (I have verified this on my local machine), if you want to run that instead.

We have identified a fix that will allow hierarchical mapping to successfully process your data and are working to get that deployed in our next release by the end of the year. The release will be announced in this community forum when it is ready. Thank you for your patience, and sorry for the inconvenience.

Details:
While the current version of hierarchical mapping does not require all of our designated marker genes to be present in the dataset, it does require there to be some marker genes at each level of the taxonomy. Unfortunately, your dataset contains zero marker genes for a few of the nodes at the “class” level of our taxonomy. As I said: there is a fix coming in the near future that will allow hierarchical mapping to proceed in spite of this.

Thank you for your interest in MapMyCells,

Scott Daniel

1 Like

Apologies, @pwashin6, I got so caught up in making the code work that I did not pay close attention to what you were saying.

With only 10 genes, I am not sure you are going to get much leverage out of hierarchical mapping, anyway. As it walks down the taxonomy tree, hierarchical mapping takes the subset of genes that are markers for that level of the tree and then does a series of searches using a random 90% of those marker genes. If there are only 10 genes total in the dataset, these searches will be using 90% of something less than 10, which will have interesting behavior.

I think the safest bet for you is to use correlation search, anyway. Correlation should give somewhat reasonable results, even with a very small number of genes.

This is just my gut assessment. We did not do testing with datasets containing fewer than a few hundred genes.

Thanks so much! The correlation search worked. Appreciate the feedback.