Tensor function Error

[hdesale2408]

Hello,

I used TCA to identify cell-type significantly differentially methylated CpGs between two exposure groups (approximately 32 in each group). In order to visualize the spread of data, I would like to present the methylation for each sample in either exposure group. Ideally, results should be graphically presented with a box plot or such for each significant CpG. I am trying to do this using the "tensor" function; however, I am getting an error.

In order to generate my model, I used the following code:

tca.mdl.chagas.2 <- tca(X = beta.values, W = out.l, C1 = pheno.tca, C2 = refactor.mdl.chagas$scores)

where:

• beta.values is a 715343 x 63 matrix containing the beta value outputs from 63 samples run on the EPIC array parsing 715343 CpG sites
• out.l is a 63 x 6 matrix containing the estimated cell-type proportions for each of the 63 samples. 6 cell types were estimated
• pheno.tca is 63 x 1 dataframe designating which exposure group each sample corresponds to
• refactor.mdl.chagas$scores is a 63 x 6 containing the results of the ReFactor adjustment

In order to find my tensor estimates I used:

tensor.mdl.hit <- tensor(tca.mdl = tca.mdl.chagas.2, X = beta.values[hits.joint,,drop=F])

where:

• tca.mdl.chagas.2 is the resulting list from the above TCA call
• beta.values[hits.joint,,drop=F]) are just the beta values associated with the significant CpGs found during a joint test of significance

the error I receive is:

INFO [2022-07-05 20:19:39] Validating input... INFO [2022-07-05 20:19:39] Starting tensor for estimating Z... INFO [2022-07-05 20:19:39] Estimate tensor... Error in X - tcrossprod(C2, deltas_hat) : non-conformable arrays

[E-R]

The tca.mdl argument is "The value returned by applying the function tca to X." (per the documentation of tensor). In your case X is a subset of the matrix which was used for fitting the model in tca.mdl.chagas.2so the dimensions don't match. You can either learn the tensor for the full data matrix and then subset the CpGs you're interested in, or alternatively, you can learn a new model via tca for the CpGs in hits.joint only and then provide that model to your existing tensor call.

[hdesale2408]

Thank you! In the resulting tensor.mdl.hit list, are the elements then referring to the estimated cell types?

[E-R]

sorry, I missed this somehow. Per the documentation,tensor.mdl.hit will be the estimated cell-type level methylation.