Viash Hub

Description

Compute a batch-balanced nearest-neighbor graph of observations on the GPU.

Wraps rapids-singlecell's rsc.pp.bbknn. For each cell it finds the nearest
neighbors within every batch separately and then merges them, correcting for
batch effects in the neighbor graph. The resulting graph is the basis for
downstream embedding and clustering (e.g. UMAP, Leiden) and is stored in the
same .uns/.obsp slots as a regular neighbor graph.

Inputs

Name	Type & Properties	Description
--input -i	file required	Input h5mu file.
--modality	string	Which modality from the input MuData file to process.
--obsm_input	string	Which .obsm slot to use as a starting embedding (use_rep).
--batch_key	string required	Which .obs column holds the batch assignment. Neighbors are searched within each batch separately and then merged.

Name

Type & Properties

--input

-i

file

required

--modality

string

--obsm_input

string

--batch_key

string

required

Outputs

Name	Type & Properties	Description
--output -o	file required output	Output h5mu file containing the found neighbors.
--uns_output	string	In which .uns slot to store the neighbor graph metadata.
--obsp_distances	string	In which .obsp slot to store the distance matrix between the resulting neighbors.
--obsp_connectivities	string	In which .obsp slot to store the connectivities matrix between the resulting neighbors.
--overwrite	boolean	Allow overwriting the .uns/.obsp output slots if they already exist.
--output_compression	string	Compression format to use for the output AnnData and/or Mudata objects. By default no compression is applied.

Name

Type & Properties

--output

-o

file

required

output

--uns_output

string

--obsp_distances

string

--obsp_connectivities

string

--overwrite

boolean

--output_compression

string

Options

Name	Type & Properties	Description
--neighbors_within_batch	integer	How many top neighbors to report for each batch. Total number of neighbors per cell is this value times the number of batches.
--n_pcs	integer	Number of dimensions of the `--obsm_input` representation to use. If not set, all available dimensions are used.
--metric	string	The distance metric to use.
--algorithm	string	The KNN query algorithm to use (provided by cuVS). See https://docs.rapids.ai/api/cuvs/stable/ for details. - brute: Brute-force exact search computing all pairwise distances. Exact but the most expensive; best for smaller datasets. - cagra: GPU graph-based approximate search optimized for high query throughput on large datasets. - ivfflat: Inverted-file index that clusters vectors into lists and only searches the nearest lists. Approximate, faster than brute. - ivfpq: Inverted-file index combined with product quantization of the vectors. Approximate with a low memory footprint; scales to very large datasets. - mg_ivfflat: Multi-GPU variant of ivfflat. - mg_ivfpq: Multi-GPU variant of ivfpq.
--trim	integer	Trim the neighbors of each cell to these many top connectivities. May help with population independence and improve the tidiness of clustering. If not set, defaults to ten times the total number of neighbors per cell.
--random_state	integer	A random seed.

Name

Type & Properties

--neighbors_within_batch

integer

--n_pcs

integer

--metric

string

--algorithm

string

--trim

integer

--random_state

integer

Run this component

Run the following command to execute this component with Nextflow:

cat > params.yaml <<'EOM'  
modality: [ "rna" ]  
obsm_input: [ "X_pca" ]  
output: "$id.$key.output"  
uns_output: [ "neighbors" ]  
obsp_distances: [ "distances" ]  
obsp_connectivities: [ "connectivities" ]  
overwrite: [ false ]  
neighbors_within_batch: [ 3 ]  
metric: [ "euclidean" ]  
algorithm: [ "brute" ]  
random_state: [ 0 ]  
id: "run"  
publish_dir: "output/"  
EOM

nextflow run https://packages.viash-hub.com/vsh/openpipeline_rapids.git \  
  -revision v0.1.3 \  
  -main-script target/nextflow/preprocessing/bbknn/main.nf \  
  -params-file params.yaml

Relationships

Used by

1 relationships

wrappers/preprocessing/bbknnopenpipeline_rapids v0.1.3

Current component

preprocessing/bbknnopenpipeline_rapids v0.1.3

Uses

0 relationships

No component dependencies found.