CIDR imputed data about cidr

[hojwk]

Hi Beata,

CIDR implements what we call 'imputed imputation', which means we never explicitly perform imputation on the gene expression value in the traditional sense. The 'imputation' was only performed inside the calculation of the cell-to-cell dissimilarity matrix, and the 'imputed' value for each dropout candidate may vary depending on which cell it is compared against. Therefore it is not possible generate an imputed gene expression matrix.

Regards,
Joshua

[JurijsNazarovs]

Hello,

What would be the best way to get the imputed matrix for my own analysis? I am asking about your cell-to-cell dissimilarity imputed matrix. Cannot see it as an element of your class.

Jurijs

[hojwk]

Hi Jurijs,

As I explained in the previous answer, we performed 'implicit imputation' to modify the calculation of cell-to-cell dissimilarity. We do not use the dissimilarity to impute the expression matrix. Therefore it is not possible to extract an imputed matrix from CIDR.

Regards,
Joshua

[JurijsNazarovs]

Joshua,

Maybe I understood you wrong, but I thought you were referring to an imputed gene-cell matrix, which is like original matrix, but with imputed values.

Meanwhile I am asking about your dissimilarity cell-to-cell imputed matrix, when you change your "o" with "o hat". Do not you provide an access to this matrix either? That sounds not right to me, since you are doing PCA and clustering on this matrix.

Do I understand you wrong?

Jurijs

[JurijsNazarovs]

Ok, I think I found an answer for my question. Please, correct me if this is wrong, but to get an imputed dissimilarity matrix, the right command is "@ dissim".

Jurijs

[hojwk]

Ah, I understand now.

The actual dissimilarity matrix is stored in the sData object. You can extract this matrix like this:

slot(sData, "dissim")

Regards,
Joshua

[JurijsNazarovs]

Joshua,

Thank you for the response. Just to make sure, this is already an imputed dissimilarity matrix. Correct?

Jurijs

[hojwk]

Yes, if it is extracted after this step:
sData <- scDissim(sData)

[JurijsNazarovs]

Great. Thank you!