failed prediciton about somaticseq HOT 3 CLOSED

It's complaining about memory....

Just a thought, you may try the xgboost algorithm in the latest version. XGBoost is orders of magnitudes faster and uses less memory than the original AdaBoost implemented in R.

So if you install the latest SomaticSeq, you can train for xgboost model from your labeled TSV files (yes, it can take multiple tsv files and will combine them automatically):
somatic_xgboost.py train -tsvs sample_1/Ensemble.sSNV.tsv sample_2/Ensemble.sSNV.tsv.... -out SNV.xgboost.classifier -threads 14

To use that classifier on new data:
somatic_xgboost.py predict -tsv Ensemble.sSNV.tsv -model SNV.xgboost.classifier -out Predicted.sSNV.tsv

To see different CLI options: somatic_xgboost.py train -h or somatic_xgboost.py predict -h.

Then, if you want to convert that into VCF file:
SSeq_tsv2vcf.py -tsv Predicted.sSNV.tsv -vcf Predicted.sSNV.vcf -tools [FILL IN ALL THE TOOLS YOU USED] -all -phred -paired.

from somaticseq.

For xgboost training, you may also pass any of the xgboost training parameters as described here by passing them into the program using this format: --extra-params max_leaves:8 max_bin:16, etc.

from somaticseq.

Hi, thanks for the suggestion. For now I increased the memory allocation and the first failed sample completed. Eventually, I will try xgboost as well. Thanks

from somaticseq.