Fig. 3.

Cosine similarity between the matching pairs of ICGC signature and signature learned by NBNMF/SNBNMF. ICGC signatures that are better matched in the signature set learned by NBNMF are on the left side of the black dashed line, and signatures that are better matched in the signature set learned by SNBNMF are on the right. The signatures are sorted in the order that the difference between the cosine similarities of the ICGC-SNBNMF signature pair and the ICGC-NBNMF signature pair increases. The gray-dotted line indicates 0.65 cosine similarity, which we use as a threshold for good matching pairs. Most of the signatures that are much better matched using NBNMF, such as signature 37, 34 and 16 has no proposed etiology by COSMIC; while most of the signatures that are much better matched by SNBNMF, such as 9, 4, 3, 6, 35, 33, 11 and 36 have certain proposed etiology