Questions about online learning

[Ping-lin14]

Hi I have a few questions when imitating the Iterative single-cell multi-omic integration using online learning tutorial to test.

1.When I run the createLiger function, I will receive an error message

pbmcs = createLiger(list(rna1 = "C:/Users/ping/Desktop/ATAC/liger_test/RNA/2018111910k_pbmc/pbmc_10k_v3_raw_feature_bc_matrix.h5",
                         rna2 = "C:/Users/ping/Desktop/ATAC/liger_test/RNA/pbmc_5k_v1/connect_5k_pbmc_NGSC3_ch1_raw_feature_bc_matrix.h5"))

Error in rbind(deparse.level, ...) : 
  numbers of columns of arguments do not match

, sometimes it will cause R to crash. Is there something wrong with my input?

2.As mentioned in the tutorial, you need to enter the raw hdf5 file to create the Liger object. Can I merge files from different platforms? For example, scRNA+scATAC. Are the codes the same?

[cgao90]

For the first issue, it is because at this moment online liger takes two HDF5 inputs that have the same set of genes/features. It is likely the two datasets used here have different sets of genes. In addition, the filtered_feature_bc_matrix are recommended, instead of the raw ones.

As for the second question: yes, online liger is able to align datasets from different platforms. As long as the slots (in HDF5 files) that store the data are correctly specified, the code should work.

Hope it helps. Thank you for your interest in trying out the online liger. Let us know if you run into any other issues.

[Ping-lin14]

Thank you very much for your answers. The problem with thecreateLiger function has been solved.

However, I encountered the second problem. When I use ATAC and RNA for online liger, I receive a warning message at the selectGenes step

Warning in selectGenes(pbmcs, var.thresh = 0.2, do.plot = F) :
  No genes were selected; lower var.thresh values or choose 'union' for combine parameter

This caused scaleNotCenter to fail. Do you have any suggestions? This is my code:

pbmcs = createLiger(list(ATAC = "C:/Users/ping/Desktop/ATAC/liger_test/ATAC/atac_pbmc_10k_nextgem_filtered_peak_bc_matrix.h5",
                         RNA = "C:/Users/ping/Desktop/ATAC/liger_test/RNA/5k_pbmc_v3_nextgem_filtered_feature_bc_matrix.h5"))

pbmcs = normalize(pbmcs)
pbmcs = selectGenes(pbmcs, var.thresh = 0.2, do.plot = F)
stim
  |==========================================================================================================================| 100%
ctrl
  |==========================================================================================================================| 100%
Warning in selectGenes(pbmcs, var.thresh = 0.2, do.plot = F) :
  No genes were selected; lower var.thresh values or choose 'union' for combine parameter

[cgao90]

Hi @Ping-lin14,

Could you try lower the var.thresh and see if there are more variable genes getting selected? This parameter has to be manually tuned every time when different datasets are used.

[jw156605]

Are you trying to select variable genes from an snATAC dataset? This will not work well. Use the datasets.use parameter to select genes from only the RNA dataset(s).

[Ping-lin14]

Hi @cgao90

I tried to reduce var.thresh to 0.01, var.genes is still 0.

[Ping-lin14]

Hi @jw156605

Thank you for your comments. I set datasets.use to 2, but it doesn't seem to work. Or is my setting wrong?

[cgao90]

I just noticed that you were using the ATAC peak by cell matrix. However, a gene-by-cell matrix is required as input. Please check out our published LIGER protocol for more details about scATAC-seq data processing as well as scRNA-seq and scATAC-seq data integration (link: https://www.nature.com/articles/s41596-020-0391-8)

[erzakiev]

Dumb question but I don't want to open an issue for that but I want to clarify, the meaning of term "online" in "online_iNMF". Is it online as in "connected to the internet and interacting with some online database online" or as in "learning features of the locally stored data in real time, in small batches"?