maotian06
commented
1 year ago If we want to build it, we can create a new workflow and write processes for each tool, probably ~ 10 processes. Some of the tools, like SnpSift and FPfilter, we have in this or other pipelines; some machine learning tools, we need to build from scratch.
But it would be worth doing so! Our current pipeline's tumor-only model is based on mutect2 and has a lot of false positives. Here is a previous result:
- Comparison between tumor-only and paired Mutect2
- sample: CPCG0196
- vcf comparison result: /hot/software/pipeline/pipeline-call-sSNV/Nextflow/development/4.0.0-rc.1/maotian-add-germline-resource/CPCG_vcf_compare_paired_tumor_only.txt
- sample: CPCG0196
| Option | Total SNPs | %Unique SNPs | %Shared SNPs |
|---|---|---|---|
| Paired Normal-Tumor | 34,747 | 14138;40.7% | 20,609; 59.3% |
| Tumor-only | 149,053 | 128,444; 86.2% | 20,609; 13.8% |
Originally posted by @maotian06 in https://github.com/uclahs-cds/pipeline-call-sSNV/discussions/80
There's a new, promising method for tumor-only SNV calling: Repository Manuscript
It looks like there are a fair number of downstream tools used/required, so it's not clear how easy this will be.