Hi,

In the SCOT v2 code, I could not self-tune the model. The align method does not have any parameter that allows self-tuning. However paper itself says that the self-tuning can be done with SCOT v2.

Can you help with this issue, please?
Thank you

Hi,

In SCOT v1 line 152 X_aligned, y_aligned= self.unsupervised_scot() we are expecting two returned parameters but the function unsupervised_scot() returns only one value. That is why SCOT v1 cannot do self-tuning how do we solve this issue?

Whilst upgrading I get the following error integrating datasets with different numbers of samples (3293 and 3164)

Traceback (most recent call last):
  File "scot_v2_try.py", line 22, in <module>
    aligned_X, aligned_y= scot_aligner.align(k=k, e=e, normalize=normalize)
  File "SCOTv2/src/scot.py", line 173, in align
    X_aligned, y_aligned = self.barycentric_projection(XontoY=XontoY)
  File "SCOTv2/src/scot.py", line 143, in barycentric_projection
    self.X_aligned=np.matmul(self.coupling, self.y) / weights[:, None]
ValueError: operands could not be broadcast together with shapes (3293,1300) (3164,1)

Here's my code

scot_aligner=SCOT(X, y)
aligned_X, aligned_y= scot_aligner.align(k=k, e=e, normalize=normalize)

Please let me know what I can change. The examples I am able to find seem to all be for the same number of samples

Hi,

I've been trying to work with the co-embedding version of scot V2, and I think there are a few typos in the coembed_datasets() method of the scotv2 class.
Once corrected, its seems to work fine, but I'd prefer if you could tell me if the changes I made in the corresponding pull request are legit?

RESOLVES #9

Thanks for your help and for your very nice work

Barthelemy Caron, postdoc in Ivan Costa's lab at RWTH Aachen.

As far as I understood preprocessing steps for snare-seq are
atac-seq dataset-> cistopic -> unit normalization
rna-seq dataset-> unit normalization -> PCA-10 components

Is that correct?

I know this is a general machine learning question, but what did you use to choose the number of components when doing PCA for a different dataset? Which tool/settings do you recommend?

The XontoY parameter seems to be missing in the align function although it was referenced in the code

Similary, the next line in the hyperparameterTuning_example.py shown below also probably throws an issue because it also seems to be missing the same hyperparameter in the function:

It seems there are only 4000 labels in splatter_labels.csv for 5000 cells in splatter_X.npy and splatter_Y.npy. I realize that I could recreate the simulations (splatter + random projections), but was hoping I could reuse your simulations. Thanks.

Hi there,

First thanks for a great method and paper. I've recently been looking at integrating scATAC-seq and scRNA-seq datasets for identically prepared but distinct cell populations. In doing so, one issue that I am wary of is that cell populations may not be represented in identical proportions in both datasets due to sampling or batch-to-batch variation in abundance. As a result, using balanced transport can result in erroneous assignment of mass between domains.

To address this, I've experimented with using an unbalanced Gromov-Wasserstein formulation and demonstrated some examples of this in two notebooks (one simulated dataset and also in SNARE-seq example). I have implemented this in a fork of the repo here: https://github.com/zsteve/SCOT

Although these modifications were motivated by my own use case, I thought this extension might be useful more generally as part of the package. If you are interested, let me know and I can make a pull request.

Stephen

Hi,

I'm hoping to reproduce the results of SCOTv2 on the MEC and RNA-Imaging datasets. I have a couple of questions:

(1) Where can I gain access to the MEC dataset (there were no links in the paper) and how did you preprocess this data? Was it conducted in the same way as SNARE-seq and scGEM?
(2) I tried replicating the results of the RNA-imaging dataset using SCOTv2's default parameters but was not able to achieve good metrics. Can you share the specific hyperparameters you used for this experiment? Additionally, was there any additional preprocessing that you took after the data loaders provided by https://github.com/uhlerlab/cross-modal-autoencoders/tree/master or did you directly use those tensor representations? For instance, did you do any of the same preprocessing steps you took in SNARE-seq/scGEM for the gene expression profiles of this dataset?

Thank you in advance for your guidance.

Hi, I wonder if there is requirements.txt in this repo, or we should create it by ourselves

An error occurs for sklearn when installing.

sklearn is deprecated as package and should be updated to scikit-learn