DeepPV is a computational pipeline that uses a deep neural network to identify somatic phased variants (PVs) from next-generation sequencing data of cell-free DNA.

How to run DeepPV

deeppv -i plasma.bam \
       -n normal.bam \
       -s selector.bed \
       -b background_directory \
       -g gnomAD_directory \
       -r repeat_directory \
       -o output_directory

How DeepPV works

Step 1: scan for PV candidates
Step 2: generate a image (18 channels) for each PV cancidate
Step 3: calculate probability based on the model
Step 4: output the PVs with the probability ≥ 0.5

Here is the snapshot of one image example:

Acknowledgements

DeepPV makes use of many open source packages:

• numpy
• pysam
• pytorch