/Concod

A framework for calling deletions

Primary LanguageShellMIT LicenseMIT

0. Create candidate set

We use Pindel, SVseq2, BreakDancer and DELLY to get raw callset and merge it as candidates.

Per line of deletions: chromosom, start, end+1

1. Feature collection

1.1 Usage

  • Need bai for each bam file, and samtools in $PATH
  • Output finename is bam_filename_normalized and bam_filename_absolute. The former is used for training model.
  • Output format: feature1, feature2.......
  • For example: ./Concod -e 0.5 -m 1000 -b test.list

2. Training and Prediction

We use LIBSVM for training and testing SVM model.

2.1 Training

  1. Add label for feature as training_data and then formatDataToLibsvm.
  2. Find the Optimal parameters and train model: python easy.py training_dataand svm-train.
  3. There are two demo model: lowCov_model and hignCov_model

2.2 Classification

  1. svm-predict testing_data.scale model predictResult
  2. Use the deletions of label with "1" as the final results.