Subset using a list of barcodes

[yoavhadas]

Hello,
Is it possible to subset a pegasus h5ad file using a list of barcodes?

Thanks,
Yoav.

[yihming]

Hello @yoavhadas ,

Sure, you can do it in a way similar to AnnData's slicing:

import pegasus as pg

pdata = pg.read_input("<your-file-name>.h5ad")
pdata_subset = pdata[barcode_list, :].copy()

where copy() is added if you want a new pegasus UnimodalData object to be created. If dropping copy(), you'll get a View of the original data object instead.

[yoavhadas]

Hi @yihming , I could subset the object using the above approach. Now I must reanalyze it (repeat the PCA, harmony, clustering, and UMAP generation) to get an updated object. When I use a similar script as I did for the original dataset, I get an error:
Traceback (most recent call last):

 File "pegasus_recluster.py", line 44, in <module>
    pegasus.neighbors(object, rep=pca_key, n_jobs = n_jobs)
  File "/sc/arion/work/tmp/miniconda3/envs/pec_pegasusenv/lib/python3.8/site-packages/pegasus/tools/nearest_neighbors.py", line 308, in neighbors
    W = calculate_affinity_matrix(indices[:, 0 : K - 1], distances[:, 0 : K - 1])
  File "/sc/arion/work/tmp/miniconda3/envs/pec_pegasusenv/lib/python3.8/site-packages/pegasusio/decorators.py", line 12, in wrapper_timer
    result = func(*args, **kwargs)
  File "/sc/arion/work/tmp/miniconda3/envs/pec_pegasusenv/lib/python3.8/site-packages/pegasus/tools/nearest_neighbors.py", line 207, in calculate_affinity_matrix
    numers = 2.0 * sigmas[i] * sigmas[indices[i, :]]
IndexError: index 228336 is out of bounds for axis 0 with size 228134

Would you happen to have any suggestions on how to overcome this issue?

Thanks
Yoav

[yihming]

Hi @yoavhadas ,

Sorry for getting back to you late. Have you tried setting use_cache=False in the pegasus.neighbors() function?

[x1han]

Hi, I have a question when working with multimodaldata, for example, I have used three matrices containing the same cells to generate one multimodaldata, as follows:

MultimodalData object with 3 UnimodalData: 'RNA-RNA', 'RNA-integrated', 'RNA-AN'
    It currently binds to UnimodalData object RNA-AN

When I filter the multimodaldata through barcode, I got a unimodaldata that only contains one matrix like this:

UnimodalData object with n_obs x n_vars = 120 x 2215
    UID: RNA-RNA; Genome: RNA; Modality: RNA
    It contains 1 matrix: 'counts'
    It currently binds to matrix 'counts' as X 

I was wondering if there is a method to subset the entire multimodaldata, allowing me to obtain multimodaldata with three matrices that only contain subset cells?

Additionally, I am not sure if you will update the pegasusIO related tutorials, as I feel the content is somewhat limited. For instance, processes like creating multimodal data from a matrix, such as CreateSeurat, took me a long time to figure out on my own.

Regardless, Pegasus is still a single-cell analysis tool that I really like. I have already used it into my own work and cited it, hoping it will continue to improve.