Identifying New Signatures Present in my Cohort

[gretherv]

Hi! I'm trying to use the Denovo Signatures feature from Musical to find new signatures in my cohort. I'm interested in identifying signatures that do not decompose well into the known COSMIC signatures.

I followed the example_full_pipeline notebook. I understand how to run denovo signature discovery, as well as matching and refitting to the COSMIC catalog.

My question is– how do I find what proportion of each "denovo" signature was decomposed into a COSMIC signature? I'm envisioning generating a plot similar to figure 3.A from your paper, where I can see how well a denovo signatures matches one of the refitted COSMIC signatures.

Secondly, how can I identify signatures that cannot be refit with high confidence to the known COSMIC signatures? In essence, how do I identify a mutational signature that might be unique to my cohort?

Thank you in advance for all of your help!

[Hu-JIN]

Hi, sorry for the late response here and for the lack of documentation for the relevant information.

Once you've obtained the de novo signatures, you can "match" the de novo signatures to a set of known signatures from the catalog using musical.refit.match(), see

MuSiCal/musical/refit.py

Line 159 in a2d8c58

def match(W, W_catalog, thresh_new_sig=0.0, method='likelihood_bidirectional', thresh=None,

. This function provides three outputs: W_s, sig_map, model. W_s is simply the set of matched known signatures from the catalog. sig_map provides the proportions you referred to, i.e., the heatmap in Fig3A of our paper. model is an instance of the SparseNNLS class, and provides more information. For example, model.H_reduced is essentially the same as sig_map, model.cos_similarities provides the cosime similarity between the original de novo signature and the reconstructed signature using the matched signatures. Therefore, to identify a novel signature, you can look at model.cos_similarities to search for de novo signatures with a low cosine similarity, indicating that known signatures in the catalog could not reconstruct the de novo signature well.

Hope this helps!

[gretherv]

Yes, that's exactly what I was looking for. Thank you for your help!