subsetting anndata on basis of louvain clusters

Question

I want to subset anndata on basis of clusters, but i am not able to understand how to do it.

I am running scVelo pipeline, and in that i ran tl.louvain function to cluster cells on basis of louvain. I got around 32 clusters, of which cluster 2 and 4 is of my interest, and i have to run the pipeline further on these clusters only. (Initially i had the loom file which i read in scVelo, so i have now the anndata.)

I tried using adata.obs["louvain"] which gave me the cluster information, but i need to write a new anndata with only 2 clusters and process further.

Please help on how to subset anndata. Any help is highly appreciated. (Being very new to it, i am finding it difficult to get)

Answer 1

If your adata.obs has a "louvain" column that I'd expect after running tl.louvain, you could do the subsetting as adata[adata.obs["louvain"] == "2"] if you want to obtain one cluster and adata[adata.obs['louvain'].isin(['2', '4'])] for obtaining cluster 2 & 4.

Answer 2

Feel free to use this function I wrote for my work.

import AnnData
import numpy as np

def cluster_sampled(adata: AnnData, clusters: list, n_samples: int) -> AnnData:
    """Sample n_samples randomly from each louvain cluster from the provided clusters

    Parameters
    ----------
    adata
        AnnData object
    clusters
        List of clusters to sample from
    n_samples
        Number of samples to take from each cluster

    Returns
    -------
    AnnData
        Annotated data matrix with sampled cells from the clusters
    """
    l = []
    adata_cluster_sampled = adata[adata.obs["louvain"].isin(clusters), :].copy()
    for k, v in adata_cluster_sampled.obs.groupby("louvain").indices.items():
        l.append(np.random.choice(v, n_samples, replace=False))
    return adata_cluster_sampled[np.concatenate(l)]