add cross-atac to snare norebook

[katosh]

I added the diffusion based atac imputation we talked about to the snare test notebook. I left the knn-based approach in there as a comparison. Since it also uses the neighborhood in RNA-expression space, the results are somewhat similar. In both cases, however, the regularization is quite strong (300 neighbors for the knn and 10 iterations for the diffusion based). It seems the 300 might have been chosen since it works well in the gene based correlations and I tried to produce comparable results with my 10 iterations.

@qinqian Please let me know if there is anything else I can do and thank you for all the organization and preparation!

[katosh]

Note that I apply the same kernel to the RNA expression before correlating it to the imputed ATAC signal. This is the analogy to the kin-kernel that was applied to ATAC and RNA in the previous example.

[qinqian]

Thank you so much @katosh for your effort, I'll merge your codes into the workflow.

@katosh, another point to explore is to use joint cell clustering of the multiomics data for aggregating cells of ATAC-seq and RNA-seq for evaluation of regulatory effect. Please let me know if you would like to try this point.

[katosh]

Hi @qinqian, I agree that theoretically a joint clustering should improve this imputation approach. However, due to the different nature (scaling, sparsity) of the modalities (RNA and ATAC) I find it non-trivial to generally unite the two without drowning one of the two modalities in the noise of the other. I will continue to consider this in my work though and let you know when I find something I can share.