File size differences of AnnData .h5ad

[alikhuseynov]

Dear Paul, I found SeuratDisk very useful, thank you for implementing it! I converted .h5ad file (~23G) to h5seurat(~28G) to work in Seurat. Then I saved it as h5seurat (becomes ~5G) and converted to .h5ad(~3.7G) and loaded it to Scanpy. questions:

• the file size of converted .h5ad is small (~3.7G), if I load it to Scanpy and safe it as .h5ad it becomes larger (~44g). Any ideas why? Probably the way Scanpy write the anndata?
• also if @metadata in Seurat object has factors, numeric, etc.. they are all transformed to integers once loaded to Scanpy as .h5ad. Any reason for that? Many Thanks Alik

[LustigePerson]

Have you checked the compressionparameter from anndata.write_h5ad()? It's turned off by default. Maybe seurat is compressing by default?

[alikhuseynov]

I don't use Scanpy much.. it might be that SeuratDisk compresses it. Thanks