diff --git a/experiments/AIDO.Cell/cell_utils.py b/experiments/AIDO.Cell/cell_utils.py index 176f592..dfa4d62 100644 --- a/experiments/AIDO.Cell/cell_utils.py +++ b/experiments/AIDO.Cell/cell_utils.py @@ -28,7 +28,8 @@ def align_adata(adata: ad.AnnData) -> Tuple[ad.AnnData, np.ndarray]: print(f'{len(new_missing_genes)} in your data that cannot be used by AIDO.Cell. Removing these.') print(new_missing_genes) print(f'{len(missing_genes)} genes in the AIDO.Cell pretraining set missing in your data.') - print('AIDO.Cell is trained with zero-masking. Setting these to zero for AIDO.Cell to ignore.') + print('Setting unknown genes to 0.') + # Todo: AIDO.Cell is trained with -1 as mask token, but read depth features don't work with -1 sum for read counts print(missing_genes) adata_missing = ad.AnnData(np.zeros((adata.shape[0], len(missing_genes)))) adata_missing.var.index = missing_genes