Mercurial > repos > goeckslab > scimap_mcmicro_to_anndata
view scimap_spatial.py @ 4:49c178651194 draft default tip
planemo upload for repository https://github.com/goeckslab/tools-mti/tree/main/tools/scimap commit c39cea5ae2ebf61b3d51d687fd9d2930c907b72d
author | goeckslab |
---|---|
date | Tue, 30 Jul 2024 18:20:49 +0000 |
parents | 8c55377d7f06 |
children |
line wrap: on
line source
import argparse import json import warnings import pandas as pd import scimap as sm from anndata import read_h5ad def main(inputs, anndata, output): """ Parameter --------- inputs : str File path to galaxy tool parameter. anndata : str File path to anndata containing phenotyping info. output : str File path to output. """ warnings.simplefilter('ignore') with open(inputs, 'r') as param_handler: params = json.load(param_handler) adata = read_h5ad(anndata) tool = params['analyses']['selected_tool'] tool_func = getattr(sm.tl, tool) options = params['analyses']['options'] # tool specific pre-processing if tool == 'cluster': options['method'] = params['analyses']['method'] subset_genes = options.pop('subset_genes') if subset_genes: options['subset_genes'] = \ [x.strip() for x in subset_genes.split(',')] sub_cluster_group = options.pop('sub_cluster_group') if sub_cluster_group: options['sub_cluster_group'] = \ [x.strip() for x in sub_cluster_group.split(',')] elif tool == 'spatial_lda': max_weight_assignment = options.pop('max_weight_assignment') for k, v in options.items(): if v == '': options[k] = None # tool execution tool_func(adata, **options) # spatial LDA post-processing if tool == 'spatial_lda': if max_weight_assignment: # assign cell to a motif based on maximum weight adata.uns['spatial_lda']['neighborhood_motif'] = \ adata.uns['spatial_lda'].idxmax(axis=1) # merge motif assignment into adata.obs adata.obs = pd.merge( adata.obs, adata.uns['spatial_lda']['neighborhood_motif'], left_index=True, right_index=True ) # write out LDA results as tabular files # so they're accessible to Galaxy users adata.uns['spatial_lda'].reset_index().to_csv( 'lda_weights.txt', sep='\t', index=False) adata.uns['spatial_lda_probability'].T.reset_index( names='motif').to_csv( 'lda_probabilities.txt', sep='\t', index=False) if 'spatial_lda_model' in adata.uns: adata.uns.pop('spatial_lda_model') adata.write(output) if __name__ == '__main__': aparser = argparse.ArgumentParser() aparser.add_argument("-i", "--inputs", dest="inputs", required=True) aparser.add_argument("-e", "--output", dest="output", required=True) aparser.add_argument("-a", "--anndata", dest="anndata", required=True) args = aparser.parse_args() main(args.inputs, args.anndata, args.output)