Bugs of running own codes for imputation

[HelloWorldLTY]

Hi, I tried to impute my own spatial datasets (as mouse) with the tutorial for imputation. However, it seems that I cannot impute it with a bug:

ValueError: None of AnnData.var.index found in pre-trained gene set. In case the input gene names are gene symbols, please enable `ensembl_auto_conversion`, or manually convert gene symbols to ensembl ids in the input dataset.

I check that my dataset is in gene name (here the genes name are all upper-case since I tried to use orthology genes.).

[wehos]

Sorry for the inconvenience. Our method used Ensembl id as gene index. We provided an automatic method to map gene names to ensembl id based on mygene here.

[HelloWorldLTY]

Hi, thanks. After transferring the data with this method, I meet a new bug: In this function:

pipeline.fit(train_data, # An AnnData object
            pipeline_config, # The config dictionary we created previously, optional
            split_field = 'split', #  Specify a column in .obs that contains split information
            train_split = 'train',
            valid_split = 'valid',
            batch_gene_list = batch_gene_list, # Specify genes that are measured in each batch, see previous section for more details
            device = DEVICE,
            ) 

     43 g2id = dict(zip(self.gene_list, list(range(len(self.gene_list)))))
     44 for batch in batch_gene_list:
---> 45     idx = torch.LongTensor([g2id[g] for g in batch_gene_list[batch]])
     46     self.batch_gene_mask[batch] = torch.zeros(len(g2id)).bool()
     47     self.batch_gene_mask[batch][idx] = True

KeyError: '0'

I think the reason is after transferring the gene name, there are some strange gene:

'ENSG00000137547',
  'ENSG00000120992',
  'ENSG00000187735',
  'ENSG00000047249',
  'ENSG00000023287',
  '0',
  'ENSG00000168300',
  '0-1',

[wehos]

Generally it is the same issue as here. Did you follow the tutorial? The tutorial should have automatically removed gene ids that are not in pretrained list.

[HelloWorldLTY]

Yes, I followed the tutorial but used my own datasets. The dataset I used is from tangram: https://github.com/broadinstitute/Tangram/blob/master/tutorial_tangram_with_squidpy.ipynb

I will try to remove all the genes with 0 or 0-id and then have a try🤔

[wehos]

Hello, I have updated the codes so that now it should work more smoothly. If you installed CellPLM with pip previously, please try pip install -U cellplm to update it accordingly. Thanks!