peterzzq / scdart Goto Github PK

Dear Authors,

Thank you for presenting your exceptional work. However, I have encountered an issue where the links provided for Mus_musculus.GRCm38.84.gtf and Homo_sapiens.GRCh37.82.gtf are currently not functional. Could you please verify and possibly re-establish access to these files?

Thank you for your attention to this matter!

The problem stated in the title caused an error in scDART_op.fit.
I downgraded the sklearn (scikit-learn) to ver 0.22.1 and the problem is gone.

Hi,
Thank you for this great method! I have a question: can I use sparse matrix for ATAC-seq data and gene activity matrix as input? My data is different from the conventional ATAC-seq data, and it actually has much more features. I tried to directly use the dense matrix, but doesn't work.

Yours sincerely,
Siqi Shen

I found if a dataset contains cells less than 1/4 the number of cells in the other dataset, the algorithm would train without using the smaller dataset at all. Look at line 142 for batch_size assignment in scDART.py and line 151 in train.py.

Also I am curious about the way you set bacth_size (1/4 of the larger dataset), would that be unnecessarily too large for dataset contain hundreds of thousands of cells in terms of memory usage? Will that be a concern if we fix batch_size at smaller number like 128 (this is what I saw other auto-encoder based algorithms used)?

Hi I get this error when trying to use scDart. Maybe it shuold be import scDart.loss as loss?

ModuleNotFoundError Traceback (most recent call last)
/home/chengwei/disco_500t/Projects/scmatchbench/notebooks/test_metrics.ipynb Cell 20 in <cell line: 1>()
----> 1 import scDART

File ~/miniconda3/envs/packages/lib/python3.9/site-packages/scDART/init.py:4, in
1 "scDART"
3 # read in main class
----> 4 from .scDART import scDART

File ~/miniconda3/envs/packages/lib/python3.9/site-packages/scDART/scDART.py:15, in
13 import scDART.dataset as dataset
14 import scDART.model as model
---> 15 import scDART.train as train
16 import scDART.utils as utils
17 import scDART.post_align as palign

File ~/miniconda3/envs/packages/lib/python3.9/site-packages/scDART/train.py:5, in
3 import torch.nn.functional as F
4 import numpy as np
----> 5 import loss
6 from torch.autograd import Variable
7 from torch import autograd

ModuleNotFoundError: No module named 'loss'

Hi,

Cloned git repo's name is scDART, and there is another scDART sub-folder nested under the main git repo.

When I appended the scDART/scDART for the sys, functions under the scDART could not be found.

I changed the original git repo's name to scDART_changename and appended scDART_changename/scDART for sys. All functions under the scDART class can be found.

I wonder if there is a better solution for this bug.

Thank you,
Yam

Hi Peter,

I am trying to do a benchmark on these methods. Since that there are quite a lot of parameters to choose from. Is there a guideline on how I should choose these parameters.

Could I ask how long does it take for the example data?

I saw from the paper, for 10000 cells, scDart spends about 2 min using GPU.

But I think I spent more than 1h

Hi, I found that most of the datasets including atac information are very large, and it takes me very long time to run the step generating chr-gene relation. Therefore, could you please offer me some suggestions about this problem? Or is it possible to run your model without this relation as an input? Thanks a lot!

Hello, I wonder how the Input coarse gene activity matrix is obtained ? Its shape is regions*genes.

If I use the function GeneActivity in Signac, the result gene.activities 's shape is new_genes * cells. So how to get the 'region2gene.csv' in your work? Are other R packages used?

Looking forward to your reply. Thx !

Hi Peter,
Sorry for sending a lot of questions! I hope this is the last one, but is it possible to obtain the updated gene activity matrix from the algorithm? I feel that it is more important for me comparing to the embeddings.

Yours sincerely,
Siqi Shen

Hi Peter,
I met another problem when running your example. One is that when running

import scDART.utils as utils

It shows the error "No module named 'scDART.utils'".

The second is that when running:

scDART_op = scDART.scDART(n_epochs = n_epochs, latent_dim = latent_dim,
ts = ts, use_anchor = use_anchor, use_potential = use_potential, k = 10,
reg_d = 1, reg_g = 1, reg_mmd = 1, l_dist_type = 'kl', seed = seeds[0],
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu'))

It shows:

Traceback (most recent call last):
File "", line 1, in
AttributeError: module 'scDART' has no attribute 'scDART'

The scDART should be successfully installed, since I can do "import scDART". What other possible problems do you think there are?

Yours sincerely,
Siqi Shen

Hello,
I am running into an error while loading the scDART package. I have attached the full error message below. I set up a conda environment (“newEnv”), activated it and installed all packages under setup.py from the linked github repository via command line.