The weighted gene expression matrix

[fenghuijian]

Hi! How do I extract the weighted gene expression matrix. The sam.adata.X should be the input gene expression matrix

[atarashansky]

Hello,

The data used as input to PCA is stored in sam.adata.uns['X_processed']. Note that this particular matrix has already been standardized/normalized prior to PCA.

If you would like the full expression matrix rescaled by all the weights, without standardization, then you'll have to calculate it yourself:

w = sam.adata.var['weights'].get_values()
weighted_matrix = sam.adata.X.multiply(w).tocsr()

[fenghuijian]

So, Doesn't the gene expression matrix get iterated? I think the gene expression matrix gets iterated.

[atarashansky]

The gene weights get iterated. Each iteration, they are used to rescale the original gene expression matrix prior to PCA.

[fenghuijian]

So, sam.adata.X is the weighted gene expression matrix for n-1 iterations, right? (n is the number of iterations)

[atarashansky]

sam.adata.X is the original expression matrix which gets rescaled by the weights prior to PCA. The weights and the nearest neighbor graph are the only objects getting updated each iteration.

[fenghuijian]

weighted_matrix = sam.adata.X.multiply(w).tocsr() This function is just multiplying the weights by the original expression matrix. And it isn't the weight gene expression matrix of the last iteration.

[atarashansky]

Those weights are the output of the final iteration! SAM iteratively updates the gene weights and nearest neighbor graph -- not the expression data. To get the weighted gene expression matrix used in the last iteration, simply multiply those weights into the gene expression matrix. Sorry for the confusion.

[fenghuijian]

Thank you very much for your reply