GKMSVM and deltaSVM version

It turned out I didn't use the R package. I downloaded gkmSVM-1.2 from the gkmSVM paper website. The scripts are here:

• TRAIN_GKMSVM/gkmsvm_kernel
• TRAIN_GKMSVM/gkmsvm_train
• TRAIN_GKMSVM/gkmsvm_classify

I forgot which version of deltaSVM I used. I downloaded it in May/June 2015. The scirpt is here: DELTASVM/deltasvm.pl.

Directory

• CTCF: The training and test set for train gkmSVM model. To match with the training data of GERV, the training set are from chr1-13 and the test set are from chr14-22. More details in the paper.
• ALL_ASB_SNP: The positive SNP set
• ALL_ASB_SNP_10000flank_wo_ALL_ASB_SNP_bootstrap10_size13030: The negative SNP set (10 sets) for Figure 3 right-most panel (CTCF).
• TRAIN_GKMSVM: the scripts to train gkmSVM, and the re-trained model output.
• DELTASVM: the scripts to score SNPs with deltaSVM.
• OUTPUT: the output of SNP scoring

Train gkmSVM

We have included the re-trained (June 17, 2016) model in the repository. To use it, you will need to also download (with wget) the kernerl file here(25G) and put it under TRAIN_GKMSVM

To train a new model (named mb.ctcf) and score the test set as well as all10mer (for deltasvm):

cd TRAIN_GKMSVM
python train_classify.py

Evaluate the performance of peak prediction on the test set (will get a AUC ~ 0.94, showing the training is successful):

cd TRAIN_GKMSVM
python evaluate.py

Score the SNPs

Use the weights of all 10 mers generated from trained gkmSVM model to score the variants:

cd DELTASVM
python score.py

Evaluate the performance (over 10 pairs of pos-neg set to get a confidence interval on ROC/PRC):

cd DELTASVM
Rscript evaluate.R

This will generate a ROC.pdf and PRC.pdf under OUTPUT/