Decomposition function does not decomposite the de novo signatures to cosmic global signatures.

[marshelma]

Running the function 'sigprofilerextractor', the suggested solution is 6 de novo signature

	Stability (Avg Silhouette)	Minimum Stability	Matrix Frobenius%	Mean Sample L1%	Maximum Sample L1%	Mean Sample L2%	Maximum Sample L2%	Significant Decrease of L2	Mean Sample KL	Maximum Sample KL
1	1	1	57.73%	51.11%	147.37%	50.98%	145.87%	TRUE	0.2305	2.0591
2	1	1	26.30%	28.52%	127.72%	25.53%	95.14%	TRUE	0.0984	1.5946
3	0.99	0.98	20.30%	24.68%	128.24%	20.89%	92.43%	FALSE	0.0834	1.561
4	0.94	0.9	16.82%	21.43%	121.02%	16.11%	96.94%	TRUE	0.0703	1.3847
5	0.8	0.12	15.70%	18.83%	83.93%	13.67%	62.86%	FALSE	0.0542	0.6204
6*	0.88	0.32	14.62%	16.52%	49.01%	11.80%	33.24%	TRUE	0.0425	0.2881
7	0.87	0.3	11.86%	14.77%	50.16%	10.35%	31.06%	FALSE	0.0348	0.3028
8	0.68	-0.49	11.95%	14.49%	49.32%	9.97%	34.42%	FALSE	0.0333	0.3079
9	0.78	-0.03	9.32%	12.23%	26.99%	8.27%	27.23%	TRUE	0.0252	0.1273
10	0.6	-0.02	9.84%	12.23%	27.57%	8.50%	26.69%	FALSE	0.0238	0.1286

However, in the decomposition folder, all samples are assigned to 8 signatures (signature A~ F, SBS1, SBS5). Here is the first a few line from the file 'Decomposed_Solution_Signatures_SBS96.txt'

MutationsType	SBS1	SBS5	Signature A	Signature B	Signature C	Signature D	Signature E	Signature F
A[C>A]A	0.000886	0.011998	0.001695	0.000153	0	0.000733	9.86E-05	0.002613
A[C>A]C	0.00228	0.009438	0.002864	0.003437	0.00096	0.004188	0.000455	1.77E-06
A[C>A]G	0.000177	0.00185	0.000913	0.000123	0	0.000608	0	1.93E-06
A[C>A]T	0.00128	0.006609	0.012232	0.019794	0	0.011603	0	1.31E-06
C[C>A]A	0.00186	0.010098	0.002824	0.002846	0.002816	0.016223	1.9E-05	0.024425
C[C>A]C	0.00122	0.005699	0.001623	0.002203	0.000484	0.027418	0.000285	0.000546
C[C>A]G	0.000115	0.00172	0.001503	0.000929	0.000289	0.00955	0	1.53E-06
C[C>A]T	0.00114	0.010098	0.015617	0.059091	0.000421	0.129235	0.001499	0.024569

The file 'comparison_with_global_ID_signatures.csv' shows

De novo extracted	Global NMF Signatures	L1 Error %	L2 Error %	Cosine Similarity
Signature 96-A	Signature 96-A	0	0	1
Signature 96-B	Signature 96-B	0	0	1
Signature 96-C	Signature 96-C	0	0	1
Signature 96-D	Signature 96-D	0	0	1
Signature 96-E	Signature 96-E	0	0	1
Signature 96-F	Signature 96-F	0	0	1

Why the de novo signatures are not deomposited to Cosmic SBS signatures?

[mishugeb]

Hi,
I am very sorry that I could not reach you earlier.

There are two reasons which will report new signatures not decomposing them into COSMIC signatures. Firstly, if the linear combination of the COSMIC signatures is less than cosine similarity 0.8 compared to the original de-novo signature; Secondly, if the original de-novo signature decomposes into more than four COSMIC signature. To see what actually happened, please check the "decomposition_logfile.txt" in the "Suggested_Solution/Decomposed_Solution" in your result.

Thanks,
Mishu