Fig. 3

Flowchart of analytical approach to classify candidate SVs into likely true or false positives. The subset of 35 annotations was chosen for Illumina paired-end data (fewer for PacBio and moleculo data) to reduce the number of annotations used in the model to those that we expected to be most important for clustering calls into different categories. The one-class model uses only the 4000 random sites for training, and it assumes that sites with annotations unlike most of these random sites are more likely to be SVs