Open Zjianglin opened 4 months ago
marcovarrone
commented
4 months ago Hi @Zjianglin, thank you for the issue! I will need some more information to debug the problem.
How big is the dataset and how much RAM do you have available? Would it be possible to run the code as a script rather than a Jupyter notebook? I think it will give much more informative error messages.
Zjianglin
commented
4 months ago Hi @marcovarrone , thanks for your reply.
the whole datasets is relative big (around 710k cells/spots from seven slices/samples). however, the error still occurred even if I only load single sample and filtered it into a AnnData object with n_obs × n_vars = 42222 × 21845. I tried SquareBin (bin50) as well as CellBin/ _in-silic segmented cells from my Stereo-seq spatial transcriptomics data, the error always occurred.
My machine is 96 CPUs, 1.9 TB RAM and a GPU Tesla V100S-PCIE-32GB. The OS is Linux n1 4.18.0-80.7.1.el8_0.x86_64, "Ubuntu 20.04.6 LTS".
$ free -h
total used free shared buff/cache available
Mem: 1.9Ti 864Gi 1.0Ti 4.6Gi 8.5Gi 1.1Ti
Swap: 31Gi 1.0Mi 31Gi
$ gpustat
n1 Thu May 16 17:28:05 2024 515.86.01
[0] Tesla V100S-PCIE-32GB | 35°C, 0 % | 0 / 32768 MB |
$ nvidia-smi
Thu May 16 17:28:17 2024
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.86.01 Driver Version: 515.86.01 CUDA Version: 11.7 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla V100S-PCI... Off | 00000000:25:00.0 Off | 0 |
| N/A 35C P0 25W / 250W | 0MiB / 32768MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
I also run the codes as a script from terminal, but it still crushed at the point. Here is the last output:
Monitored metric elbo_validation did not improve in the last 45 records. Best score: 1892.535. Signaling Trainer to stop.
^M 0%| | 0/4 [00:00<?, ?it/s]^M100%|██████████| 4/4 [00:00<00:00, 26.76it/s]^M100%|██████████| 4/4 [00:00<00:00, 26.73it/s]
AnnData object with n_obs × n_vars = 42222 × 21845
obs: 'dnbCount', 'area', 'nCount_Spatial', 'nFeature_Spatial', 'percent.mito', 'orig.ident', 'x', 'y', 'sample', 'n_counts', 'n_genes', '_scvi_batch', '_scvi_labels'
var: 'n_cells', 'n_counts', 'mean_umi'
uns: 'log1p', '_scvi_uuid', '_scvi_manager_uuid', 'spatial_neighbors'
obsm: 'spatial', 'X_scVI', 'X_cellcharter'
layers: 'counts'
obsp: 'spatial_connectivities', 'spatial_distances'
Iteration 1/10
^M 0%| | 0/23 [00:00<?, ?it/s]/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/pytorch_lightning/trainer/setup.py:176: PossibleUserWarning: GPU available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='gpu', devices=1)`.
rank_zero_warn(
^M 4%|▍ | 1/23 [00:07<02:52, 7.82s/it]/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/pytorch_lightning/trainer/setup.py:176: PossibleUserWarning: GPU available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='gpu', devices=1)`.
rank_zero_warn(
^M 9%|▊ | 2/23 [00:17<03:11, 9.14s/it]/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/pytorch_lightning/trainer/setup.py:176: PossibleUserWarning: GPU available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='gpu', devices=1)`.
rank_zero_warn(
Mmmh that's weird. Never happened before.
Try to run :
tl.Cluster instead of tl.ClusterAutoK.
That would be something like:
gmm = cc.tl.ClusterAutoK(
n_clusters=10,
)
gmm.fit(adata, use_rep='X_cellcharter')Hi @marcovarrone , the tl.Cluster can normally run as your tutorial when only input a small subset of samples (~80k cells).
However, when I input five samples (~510k cells), the process was killed as before.
Here is the last output:
AnnData object with n_obs × n_vars = 513226 × 24944
obs: 'dnbCount', 'area', 'nCount_Spatial', 'nFeature_Spatial', 'percent.mito', 'orig.ident', 'x', 'y', 'sample', 'n_counts', 'n_genes', '_scvi_batch', '_scvi_labels'
var: 'n_counts'
uns: 'log1p', '_scvi_uuid', '_scvi_manager_uuid', 'spatial_neighbors'
obsm: 'spatial', 'X_scVI', 'X_cellcharter'
layers: 'counts'
obsp: 'spatial_connectivities', 'spatial_distances'
/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/pytorch_lightning/trainer/setup.py:176: PossibleUserWarning: GPU available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='gpu', devices=1)`.
rank_zero_warn(
Epoch 1: 100%|██████████| 1/1 [00:00<00:00, 8.53it/s]
Epoch 34: 0%| | 0/1 [00:00<?, ?it/s, nll=16.80] Ok @Zjianglin , then I am quite sure it's a memory problem, but I clustered more than 8 million cells with less than 100 GB so there is something wrong happening here.
Some things that come to mind:
htop and top during the execution to make sure that this is the problem?obsm['spatial']? In that case the Delaunay triangulation can add too many edges, so remove the Z dimension. (I doubt this is the case otherwise the memory error would happen in aggregate neighbors)adata.obsm['X_scVI'] and adata.obsm['X_cellcharter']?Sorry if it's taking time but without access to the data and the environment it's not easy to debug.
Zjianglin
commented
4 months ago Hi @marcovarrone ,
model.train(early_stopping=True, enable_progress_bar=True) for ~510k cells, the RAM usage of the process about 43.5GB. During run gmm.fit, it seems use all CPUs but the RAM seems not too big.spatial, the size of adata.obsm['X_scVI'] and adata.obsm['X_cellcharter']
AnnData object with n_obs × n_vars = 513226 × 24944
obs: 'dnbCount', 'area', 'nCount_Spatial', 'nFeature_Spatial', 'percent.mito', 'orig.ident', 'x', 'y', 'sample', 'n_counts', 'n_genes', '_scvi_batch', '_scvi_labels'
var: 'n_counts'
uns: 'log1p', '_scvi_uuid', '_scvi_manager_uuid', 'spatial_neighbors', 'sample_colors'
obsm: 'spatial', 'X_scVI', 'X_cellcharter'
layers: 'counts'
obsp: 'spatial_connectivities', 'spatial_distances'
adata samples: ['WT', 'LowMM1479', 'LowMM1496', 'MedianLL1084', 'Median1362']
Categories (5, object): ['LowMM1479', 'LowMM1496', 'Median1362', 'MedianLL1084', 'WT']
adata.obsm['spatial']:
[[ 7677 10238]
[ 7676 10189]
[ 7673 10218]
...
[14438 14411]
[14419 14423]
[14601 14463]]adata.uns['spatial_neighbors']: {'connectivities_key': 'spatial_connectivities', 'distances_key': 'spatial_distances', 'params': {'n_neighbors': 6, 'coord_type': 'generic', 'radius': 99.02019995940222, 'transform': None}}
Size of adata.obsm['X_scVI']: (513226, 10) Size of adata.obsm['X_cellcharter']: (513226, 40)
-the results of the cluster 10 for a single sample seems reasonable, so i want to run it for all samples.
-I failed to plot the tissue sample showing the links between cells from the spatial neighbors for the error below:
```python
sq.pl.spatial_scatter(
adata,
color='sample',
library_key='sample',
title=['WT', 'LowMM1479', 'LowMM1496', 'MedianLL1084', 'Median1362'],
size=2500,
img=None,
connectivity_key='spatial_connectivities',
edges_width=0.3,
legend_loc=None,
# crop_coord=(900000, 900000, 1700000, 1700000 ),
library_id=['WT', 'LowMM1479', 'LowMM1496', 'MedianLL1084', 'Median1362']
)
#==>
#...
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/squidpy/_constants/_pkg_constants.py:181, in Key.uns._sort_haystack(cls, adata, spatial_key, library_id, sub_key)
172 @classmethod
173 def _sort_haystack(
174 cls,
(...)
178 sub_key: str | None = None,
179 ) -> Sequence[str] | None:
180 if spatial_key not in adata.uns:
--> 181 raise KeyError(f"Spatial key {spatial_key!r} not found in `adata.uns`.")
182 haystack = list(adata.uns[spatial_key])
183 if library_id is not None:
KeyError: "Spatial key 'spatial' not found in `adata.uns`."
Hi @marcovarrone , I solved this problem by resetting the CPU time of ulimit to unlimited on my server. It may due to the cpu time is limited to 1980 seconds in ulimit for the node. Thanks for your reply.
Btw, as I analyzed seven slices/samples together, I wonder if the cellcharter clustering process includes batch-correction? i.e. If I get the cellcharter clusters, could I just find DEG among the clusters on the normlaized count matrix? Or should do some other batch-correction such as Harmony or BBKNN? thanks.
marcovarrone
commented
2 months ago Hi @Zjianglin, amazing! I will add it to the documentation in case it happens to someone else :) thank you for figuring it out
since you used
scvi.model.SCVI.setup_anndata(
adata,
layer="counts",
batch_key='sample'
)there was a batch correction on the samples.
Even though scVI is quite good at not removing biological variation, as a general approach I would suggest you run scVI without batch_key='sample', i.e., without batch correction, and see if there are batch effects. If there are, then you run scVI with batch effect correction.
I know it's more work but it's important to use batch effect correction only if necessary because every batch correction method brings a certain risk of removing true biological variability.
Zjianglin
commented
2 months ago @marcovarrone Thanks for you suggestion. I will try it later.
After learning the tutorial of CosMx SRT, I have another few questions.
cc.pl.autok_stability(autok) output, you said that 5 and 11 are good candidates.. I wonder why the good candidates is 11 rather than 8? It seems 8 has the second high stability.n_layers=3 in aggregate_neighbors and used default settings of SCVI.setup_anndata. Is that your recommendation for these settings or should we tried another more or less n_layers. What's the tendency of these parameters?bin50 in my stereo-seq data), I want to do a clusters DEG identification and cell type annotation. Could I just use the scanpy.tl.rank_genes_groups with groupby="cluster_cellcharter" following the codes of the tutorial, or should I do some additional preprocessings. For the Q3, I tried to find DEG using sc.tl.rank_genes_groups(adata, groupby='cluster_cellcharter', method="wilcoxon", key_added=degkey), but it print out a warning :WARNING: It seems you use rank_genes_groups on the raw count data. Please logarithmize your data before calling rank_genes_groups. . However, the adata has been logarithmized before:
adata.layers["counts"] = adata.X.copy()
sc.pp.normalize_total(adata, target_sum=1e6)
sc.pp.log1p(adata)The pipeline did not set raw attributes of adata, but the raw attribute does exists and its value are not be logarithmized. Is it come from the result of adata.layers["counts"] = adata.X.copy() command?
print('AnnData raw is {}'.format(adata.raw))
print('AnnData raw is {}'.format(adata.raw.shape))
# check .raw whether is log1p data
print(f"{adata.X.min()}\n{adata.X.max()}")
print('Raw X min: {}, max {}'.format(adata.raw.X.min(), adata.raw.X.max()))
#==>
AnnData raw is Raw AnnData with n_obs × n_vars = 326394 × 22646
AnnData raw is (326394, 22646)
0.0
12.908868679138859
Raw X min: 0.0, max 965.0Did I do anything wrong?. The code is same as the beginning of this issue, according to you tutorial of CosMx for my stereo-seq bin50 data.
Thanks again. I'm looking forward to your reply.
marcovarrone
commented
2 months ago @Zjianglin
scVI parameters: I saw that the default settings for scVI tend to work well for all cases. In more in-depth analysis I tend to fine-tune a bit scVI to get better reconstruction error, without compromising too much the KL loss. It's not common to do it and it's if you want to squeeze the most out of that, so if you don't have experience with that or you don't feel comfortable in changing the scVI parameters, then the default values should be fine.n_layers: it's a very important parameter, because it represents how big is the neighborhood taken into account when computing the cell niches. n_layers=0 means that the clustering will be performed on the cell features only (which means clustering based on the cell type/cell state). n_layers=1 means that the clustering will be performed on the cell features, but also based on the features of the first layer of neighbors around the cell. And so on. The higher the value of n_layers the broader will be the scope of the niches, with an extremely high value leading with a single niche across the whole sample.n_layers=0), and annotate the cell types, for example using DEG. Then you can do DEF between the same cell type but that belongs to different niches.
The reason why it's giving you that warning is (I think) because the function uses the raw data by default if the raw data are present. Given that you told me that you have the raw data in the AnnData object, you just need to set the parameter use_raw=False. I didn't test it, so I am not 100% sure.It should be clear in the paper, but I repeat it here because I am starting to see this misunderstanding often: CellCharter has not been initially designed to find cell types but to find cell niches, which are areas with the same combination of cell types and cell states. You can look for cell types by running it with n_layers=0 and it could be convenient because it's very scalable, but this is not its original purpose.
Zjianglin
commented
2 months ago Hi @marcovarrone thanks for you reply. After clustering, I tried to perform Proximity analysis using cc.gr.diff_nhood_enrichment. However, it always be stuck after some time running. I have tried different n_jobs from 15 to 48, but the problem always occur.
Here is the running output:
AnnData object with n_obs × n_vars = 326394 × 19517
obs: 'total_counts', 'n_genes_by_counts', 'pct_counts_mt', 'orig.ident', 'x', 'y', 'spatial1', 'spatial2', 'n_counts', 'n_genes', 'sample', '_scvi_batch', '_scvi_labels', 'cluster_CC28', 'cluster_CC9'
uns: '_scvi_manager_uuid', '_scvi_uuid', 'log1p', 'sample_colors', 'spatial', 'spatial_neighbors'
obsm: 'X_cellcharter', 'X_scVI', 'spatial'
layers: 'counts'
obsp: 'spatial_connectivities', 'spatial_distances'
cc.gr.diff_nhood_enrichment(
adata,
cluster_key=grp_colx,
condition_key='condition',
library_key='sample',
pvalues=True,
n_jobs=36,
n_perms=100
)
#==>
adata.obs[cluster_key][np.repeat(np.arange(matrix.indptr.shape[0] - 1), np.diff(matrix.indptr))]
/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_build.py:123: FutureWarning: Series.__getitem__ treating keys as positions is deprecated. In a future version, integer keys will always be treated as labels (consistent with DataFrame behavior). To access a value by position, use `ser.iloc[pos]`
target_clusters = np.array(adata.obs[cluster_key][matrix.indices])
/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_build.py:125: FutureWarning: Series.__getitem__ treating keys as positions is deprecated. In a future version, integer keys will always be treated as labels (consistent with DataFrame behavior). To access a value by position, use `ser.iloc[pos]`
adata.obs[cluster_key][np.repeat(np.arange(matrix.indptr.shape[0] - 1), np.diff(matrix.indptr))]
2%|▏ | 2/100 [01:03<48:39, 29.79s/it] /home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_build.py:123: FutureWarning: Series.__getitem__ treating keys as positions is deprecated. In a future version, integer keys will always be treated as labels (consistent with DataFrame behavior). To access a value by position, use `ser.iloc[pos]`
target_clusters = np.array(adata.obs[cluster_key][matrix.indices])
/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_build.py:125: FutureWarning: Series.__getitem__ treating keys as positions is deprecated. In a future version, integer keys will always be treated as labels (consistent with DataFrame behavior). To access a value by position, use `ser.iloc[pos]`
# ...
target_clusters = np.array(adata.obs[cluster_key][matrix.indices])
/home/ncba04/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_build.py:125: FutureWarning: Series.__getitem__ treating keys as positions is deprecated. In a future version, integer keys will always be treated as labels (consistent with DataFrame behavior). To access a value by position, use `ser.iloc[pos]`
adata.obs[cluster_key][np.repeat(np.arange(matrix.indptr.shape[0] - 1), np.diff(matrix.indptr))]
72%|███████▏ | 72/100 [28:22:32<11:02:06, 1418.79s/it]
---------------------------------------------------------------------------
BrokenProcessPool Traceback (most recent call last)
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_nhood.py:342, in diff_nhood_enrichment(adata, cluster_key, condition_key, condition_groups, library_key, pvalues, n_perms, n_jobs, copy, **nhood_kwargs)
341 for future in as_completed(futures):
--> 342 expected_permuted = future.result()
343 expected.append(expected_permuted)
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/concurrent/futures/_base.py:451, in Future.result(self, timeout)
450 elif self._state == FINISHED:
--> 451 return self.__get_result()
453 self._condition.wait(timeout)
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/concurrent/futures/_base.py:403, in Future.__get_result(self)
402 try:
--> 403 raise self._exception
404 finally:
405 # Break a reference cycle with the exception in self._exception
BrokenProcessPool: A process in the process pool was terminated abruptly while the future was running or pending.
During handling of the above exception, another exception occurred:
KeyboardInterrupt Traceback (most recent call last)
Cell In[8], line 4
1 #Proximity analysis
2 # The next step is to describe the relative arrangement of spatial clusters in the samples and
3 # observe whether the relative proximity between spatial clusters changes between the different condition
----> 4 cc.gr.diff_nhood_enrichment(
5 adata,
6 cluster_key=grp_colx,
7 condition_key='condition',
8 library_key='sample',
9 pvalues=True,
10 n_jobs=36,
11 n_perms=100
12 )
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_nhood.py:327, in diff_nhood_enrichment(adata, cluster_key, condition_key, condition_groups, library_key, pvalues, n_perms, n_jobs, copy, **nhood_kwargs)
324 adata_perms = adata.copy()
326 with tqdm(total=n_perms) as pbar:
--> 327 with ProcessPoolExecutor(max_workers=n_jobs) as executor:
328 futures = []
330 for samples_perm in samples_perms:
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/concurrent/futures/_base.py:649, in Executor.__exit__(self, exc_type, exc_val, exc_tb)
648 def __exit__(self, exc_type, exc_val, exc_tb):
--> 649 self.shutdown(wait=True)
650 return False
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/concurrent/futures/process.py:780, in ProcessPoolExecutor.shutdown(self, wait, cancel_futures)
777 self._executor_manager_thread_wakeup.wakeup()
779 if self._executor_manager_thread is not None and wait:
--> 780 self._executor_manager_thread.join()
781 # To reduce the risk of opening too many files, remove references to
782 # objects that use file descriptors.
783 self._executor_manager_thread = None
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/threading.py:1096, in Thread.join(self, timeout)
1093 raise RuntimeError("cannot join current thread")
1095 if timeout is None:
-> 1096 self._wait_for_tstate_lock()
1097 else:
1098 # the behavior of a negative timeout isn't documented, but
1099 # historically .join(timeout=x) for x<0 has acted as if timeout=0
1100 self._wait_for_tstate_lock(timeout=max(timeout, 0))
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/threading.py:1116, in Thread._wait_for_tstate_lock(self, block, timeout)
1113 return
1115 try:
-> 1116 if lock.acquire(block, timeout):
1117 lock.release()
1118 self._stop()
KeyboardInterrupt:
The KeyboardInterrupt was done after about 24 hours running and no log change more than 10 hours.
marcovarrone
commented
2 months ago That's also unexpected @Zjianglin . How many clusters, conditions and samples do you have?
I am running gr.diff_nhood_enrichment for a dataset of 240k cells with 6 clusters and 2 conditions (one condition with 90k cells and the other one with 150k).
With n_jobs=1 it takes 2.4 seconds per iterations and with n_jobs=15 takes 1.6 seconds per iteration (there is some overhead given by the parallelization unfortunately). Try with n_jobs=1 and let's see what's the speed.
Moreover, which version of CellCharter are you using?
Zjianglin
commented
2 months ago Hi @marcovarrone I have four conditions and 7 samples/slices (one of the slices has two samples, so the total number of samples is 8 / Sample in obs ), A total of 326k bin50/cells were included for this analysis. The total cellcharter clusters were 28 based stability analysis with a second high stability.
Sample
LowMM1479 68533
HighCF142 52070
LowMM1496 51378
Median1362 37066
WT8m 30891
HighPT2203 29979
MedianLL1084 29468
WT4m 27009
Name: count, dtype: int64
sample
LowMM1479 68533
WT 57900
HighCF142 52070
LowMM1496 51378
Median1362 37066
HighPT2203 29979
MedianLL1084 29468
Name: count, dtype: int64
condition
Low 119911
High 82049
Median 66534
WT 57900
Name: count, dtype: int64
My CellCharter is 0.2.2
After setting the n_jobs=1, the function raised a Error after 100%|██████████| 100/100 [48:38<00:00, 29.19s/it]. Here is the final output:
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In[45], line 5
1 print(adata)
2 #Proximity analysis
3 # The next step is to describe the relative arrangement of spatial clusters in the samples and
4 # observe whether the relative proximity between spatial clusters changes between the different condition
----> 5 cc.gr.diff_nhood_enrichment(
6 adata,
7 cluster_key=grp_colx,
8 condition_key='condition',
9 library_key='Sample',
10 pvalues=True,
11 n_jobs=1, #36
12 n_perms=100
13 )
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/cellcharter/gr/_nhood.py:346, in diff_nhood_enrichment(adata, cluster_key, condition_key, condition_groups, library_key, pvalues, n_perms, n_jobs, copy, **nhood_kwargs)
343 expected.append(expected_permuted)
344 pbar.update(1)
--> 346 expected = np.stack(expected, axis=0)
347 tmp_pvalues = _empirical_pvalues(observed, expected)
348 result[result_key]["pvalue"] = tmp_pvalues
File ~/miniforge3/envs/cellcharter-env/lib/python3.10/site-packages/numpy/core/shape_base.py:449, in stack(arrays, axis, out, dtype, casting)
447 shapes = {arr.shape for arr in arrays}
448 if len(shapes) != 1:
--> 449 raise ValueError('all input arrays must have the same shape')
451 result_ndim = arrays[0].ndim + 1
452 axis = normalize_axis_index(axis, result_ndim)
ValueError: all input arrays must have the same shape
I tried library_key='Sample' (8 samples, each condition has two samples) as well as library_key='sample' (7 samples, WT has one sample/ two sample slices on the one chip), but the error still occurs. Could you please help me?
marcovarrone
commented
2 months ago Thanks @Zjianglin, you actually caught a bug in CellCharter!
The problem was rising whenever certain clusters were not shared between two conditions. Updating to 0.2.3 should (hopefully) resolve the problem. Let me know and thank you very much again for help testing CellCharter :)
Report
Hi developers, I'm trying to run
cellcharteron my ownStereo-Seqspatial transcriptomics data. I have seven samples and tried to runcellcharterfollowing the Spatial clustering of Nanostring CosMx spatial transcriptomics data.However, it always raise a
The kernel appears to have died. It will restart automatically.error. I have tried use a subset samples (reduced cell numbers from68w+to37w+) and useCPUinstead ofGPU, but the error still occurs. Could you please help me figure it out? Thanks.Here is my running codes:
All above codes normally executed, but always failed when running codes below:
Here is my machine info (96 processes with 1.9TB memory):
Version information
session_info 1.0.0
asttokens NA colorama 0.4.6 comm 0.2.2 cython_runtime NA dateutil 2.9.0 debugpy 1.8.1 decorator 5.1.1 exceptiongroup 1.2.0 executing 2.0.1 ipykernel 6.29.3 jedi 0.19.1 packaging 24.0 parso 0.8.4 pickleshare 0.7.5 platformdirs 4.2.1 prompt_toolkit 3.0.42 psutil 5.9.8 pure_eval 0.2.2 pydev_ipython NA pydevconsole NA pydevd 2.9.5 pydevd_file_utils NA pydevd_plugins NA pydevd_tracing NA pygments 2.18.0 six 1.16.0 stack_data 0.6.2 tornado 6.4 traitlets 5.14.3 typing_extensions NA wcwidth 0.2.13 zmq 26.0.3
IPython 8.24.0 jupyter_client 8.6.1 jupyter_core 5.7.2
Python 3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) [GCC 12.3.0] Linux-4.18.0-80.7.1.el8_0.x86_64-x86_64-with-glibc2.31