Whether or How to determine the number of signatures (K) automatically? about tcsm HOT 3 CLOSED

@WuyangFF95

In the paper and in our experiments, we plot heldout log-likelihood to determine K. I'll add that script to the demo directory but feel free to poke around in the TCGA-BRCA experiment to see.

from tcsm.

Thank you @wir963 !

I went to the demo folder, and ran help for stm_heldout_likelihood.R.

(tcsm) [wuyang@monster demo]$ ../src/stm_heldout_likelihood.R -h
usage: ../src/stm_heldout_likelihood.R [-h] [--seed SEED]
[--covariates COVARIATES]
[--trainf TRAINF] [--testf TESTF]
[--heldout HELDOUT]
trainmc testmc k

positional arguments:
trainmc mutation count input file for training
testmc mutation count input file for test
k number of signatures to use

optional arguments:
-h, --help show this help message and exit
--seed SEED random seed
--covariates COVARIATES
covariates (separated by +)
--trainf TRAINF feature input file for training
--testf TESTF feature input file for test
--heldout HELDOUT feature input file for test

But I'm still confused about a couple of fields:

1. How to split the original mutation count file into training and test set? Let's say I have 500 tumors in a mutation count file, should I put 450 tumors as training set and 50 tumors as test set, and do this for 10 times?
2. What is feature input file ? What is the difference between --testf TESTF and --heldout HELDOUT ?
3. Can you provide me an example command to run ../src/stm_heldout_likelihood.R? Thanks!

from tcsm.

@WuyangFF95,

The feature input file is required when you use covariates (it isn't necessary otherwise). I copy-pasted the help message for --heldout from --testf and forgot to fix it but it should be obvious now.

If you look at the demo/Snakefile file, you'll see a workflow for running TCSM (both for plotting the heldout likelihood and estimating signatures and exposures), which will include an example command for ../src/stm_heldout_likelihood.R

from tcsm.