query data projection error

[kaizen89]

Hi, I have a reference dataset that I integrated with simspec, now I would like to project on it a small public dataset to see where it would project on the UMAP space. Unfortunately the code that you provide on the README throws an error

model <- cluster_sim_spectrum(seu_object, label_tag = "orig.ident", return_seuratObj = F)
seurat_query <- css_project(obj_pub, model)
Error in as(Y, "dgCMatrix") : 
  no method or default for coercing “numeric” to “dgCMatrix”

Could you please provide a small example on how to proceed with the projection ? Thanks

[daniel-spies]

hi there,

thanks for the great and super fast tool!

I encountered the same error when using a model that was created from a simspec object after calling clusters and using the cluster number as the label_tag. If instead I create the model using 'orig.ident' I get another error

model <- cluster_sim_spectrum(data_normal, label_tag='orig.ident', num_pcs_compute = ndim, num_pcs_use = ndim, k=k, cluster_resolution=0.6, spectrum_type='corr_kernel', corr_method= 'pearson' ,threads=4, return_seuratObj = F)
simspec_tumor <- css_project(data_tumor@assays$RNA@data, model)
Error in sim * model$args["lambda"] : 
non-numeric argument to binary operator

in both cases the error occurs when combining the model$profiles in the ref_sim_spectrum.default section.

best Daniel

[kaizen89]

Hi @zhisonghe , could you please provide some help regarding this issue? Thanks a lot!

[zhisonghe]

Hi @kaizen89 and @daniel-spies

Thanks for using simspec and really sorry for such a long delay on response due to some personal issues.

May I ask for a bit more details about the model, particularly the number of clusters obtained during clustering on each reference sample? You can check it by looking into model$model$profiles which should be a list with length equal to the reference sample number, with each element being a matrix with columns representing the clusters of that sample. I'm wondering whether any of those matrices has only one column.

Look forward to hearing back from you.

Best, Zhisong

[kaizen89]

You are right, when I removed the sample which had one column only the error was gone. Thanks!

[zhisonghe]

In the recent update there is the min_cluster_num parameter added to the cluster_sim_spectrum function with default value of 3. This value sets a threshold on the cluster number per sample and samples with too few clusters will be discarded from the reference panel. This should essentially avoid such an issue.