Handle reading empty H5AD slots

[lazappi]

Reading currently fails if a slot is not set in the .h5ad file

[rcannood]

Can you post a minimal reproducible example? ;)

[LouiseDck]

e.g. when reading in a file with only a .X:

Error in private$.validate_layers(value) : 
  'layers' must must be a named list
In addition: Warning messages:
1: In private$.validate_obsvar_dataframe(value, "obs") :
  'obs' should not have any rownames, removing them from the data frame.
2: In private$.validate_obsvar_dataframe(value, "var") :
  'var' should not have any rownames, removing them from the data frame.``

[kollo97]

Hi! Thanks for this great tool first of all.
I saw that this bug fix is part of your milestone for version 1.0.0, I just wanted to say that I'd also appreciate a solution on the long run!
And add minimal reproducible example:
in python:

# create test anndata without var and obs
import anndata as ad
from scipy.sparse import csr_matrix
import numpy as np

counts = csr_matrix(np.random.poisson(1, size=(100, 2000)), dtype=np.float32)
adata = ad.AnnData(counts)
print(adata.var.head())
adata.write_h5ad("no_var_no_obs.h5ad")


counts = csr_matrix(np.random.poisson(1, size=(100, 2000)), dtype=np.float32)
adata = ad.AnnData(counts)
adata.var["gene"] = adata.var.index
adata.obs["cell"] = adata.obs.index
print(adata.var.head())
adata.write_h5ad("var_obs.h5ad")

Then in R:

library(anndataR)
library(Seurat)
se1 <- read_h5ad("no_var_no_obs.h5ad", to = "Seurat")  # fails
# Error in if (ncol(obj$obs) > 0) { : argument is of length zero
# In addition: Warning messages:
# 1: In value[[3L]](cond) : Error reading element 'var' of type 'dataframe':
# HDF5-API Errors:
#     error #000: H5A.c in H5Aread(): line 1043: can't synchronously read data
#         class: HDF5
#         major: Attribute
#         minor: Read failed

#     error #001: H5A.c in H5A__read_api_common(): line 1003: buf parameter can't be NULL
#         class: HDF5
#         major: Invalid arguments to routine
#         minor: Bad value

# 2: In value[[3L]](cond) : Error reading element 'obs' of type 'dataframe':
# HDF5-API Errors:
#     error #000: H5A.c in H5Aread(): line 1043: can't synchronously read data
#         class: HDF5
#         major: Attribute
#         minor: Read failed

#     error #001: H5A.c in H5A__read_api_common(): line 1003: buf parameter can't be NULL
#         class: HDF5
#         major: Invalid arguments to routine
#         minor: Bad value

# 3: In value[[3L]](cond) : Error reading element 'var' of type 'dataframe':
# HDF5-API Errors:
#     error #000: H5A.c in H5Aread(): line 1043: can't synchronously read data
#         class: HDF5
#         major: Attribute
#         minor: Read failed

#     error #001: H5A.c in H5A__read_api_common(): line 1003: buf parameter can't be NULL
#         class: HDF5
#         major: Invalid arguments to routine
#         minor: Bad value

# 4: In value[[3L]](cond) : Error reading element 'obs' of type 'dataframe':
# HDF5-API Errors:
#     error #000: H5A.c in H5Aread(): line 1043: can't synchronously read data
#         class: HDF5
#         major: Attribute
#         minor: Read failed

#     error #001: H5A.c in H5A__read_api_common(): line 1003: buf parameter can't be NULL
#         class: HDF5
#         major: Invalid arguments to routine
#         minor: Bad value

se2 <- read_h5ad("var_obs.h5ad", to = "Seurat")  # works

se3 <- read_h5ad("no_var_no_obs.h5ad")  # works but throws warning messages 
# Warning messages:
# 1: In value[[3L]](cond) : Error reading element 'obs' of type 'dataframe':
# HDF5-API Errors:
#     error #000: H5A.c in H5Aread(): line 1043: can't synchronously read data
#         class: HDF5
#         major: Attribute
#         minor: Read failed

#     error #001: H5A.c in H5A__read_api_common(): line 1003: buf parameter can't be NULL
#         class: HDF5
#         major: Invalid arguments to routine
#         minor: Bad value

# 2: In value[[3L]](cond) : Error reading element 'var' of type 'dataframe':
# HDF5-API Errors:
#     error #000: H5A.c in H5Aread(): line 1043: can't synchronously read data
#         class: HDF5
#         major: Attribute
#         minor: Read failed

#     error #001: H5A.c in H5A__read_api_common(): line 1003: buf parameter can't be NULL
#         class: HDF5
#         major: Invalid arguments to routine
#         minor: Bad value

se3 <- se3$to_Seurat()  # works

As mentioned in the python code, a quick fix is to just add the var and obs indices as columns.

[rcannood]

@lazappi @LouiseDck The original issue still persists:

import anndata as ad
ad.AnnData().write_h5ad("nothing.h5ad")

library(anndataR)
read_h5ad("nothing.h5ad")

# Error in `[[.H5File`(file, name) : 
#  An object with name X does not exist in this group

I'll try to resolve this issue asap, since it will likely be relatively easy to solve.

@kollo97 Thanks for reporting this issue. However, it seems to be unrelated to the original issue. Which version of R and the dependencies you have installed are you using? I believe you are likely using an older version of hdf5r, since the code you provided works on my end:

library(anndataR)
se1 <- read_h5ad("no_var_no_obs.h5ad", to = "Seurat")  # works
# Loading required namespace: SeuratObject
se2 <- read_h5ad("var_obs.h5ad", to = "Seurat")  # works
se3 <- read_h5ad("no_var_no_obs.h5ad") # works
se3 <- se3$to_Seurat()  # works

If the problem still persists, please create a separate issue.

[kollo97]

Hi @rcannood, thanks so much for looking into this so quickly!!
Sorry to have bothered you, the problem was on my end..
I tried the code in a fresh environment and it works indeed without throwing any HDF5-API warnings. It turns out, I only get this warning when working in a tmux session, so I must have some misconfiguration..
Thanks again 🙌

[rcannood]

@kollo97 Fantastic 👍

@ lazappi and LouiseDck In the meantime, the original issue has also been resolved:

import anndata as ad
ad.AnnData().write_h5ad("nothing.h5ad")

library(anndataR)
read_h5ad("nothing.h5ad")

# AnnData object with n_obs × n_vars = 0 × 0