If you want to assess how similar signatures in your data are to a set of known signatures, for examples the ones published by Alexandrov and colleagues, you can first estimate the signatures from the data and then compute the similarity between the two sets of signatures. The idea is discussed in another thread: SomaticSignatures: Compare my datas with validated mutational signatures. Generally, you can retrieve the estimated signatures after a fit with signatures(object).

Directly considering a set of known signatures in the matrix decomposition is not an implemented high-level feature of the SomaticSignatures package. One can utilise the low-level functionality of the package for this. Can you describe your use case in a bit more detail? Then we can see how this might be possible to compute.