Probhap: probabilistic single-individual haplotyping

Probhap is a software package for reconstructing parental haplotypes from long reads. It implements the algorithm proposed in:

V. Kuleshov, Probabilistic single-individual haplotyping, Bioinformatics (2014) 30 (17): 379-385.

Probhap works best with very long reads at a relatively shallow coverage (<= 12X). Its advantages are improved accuracy and an assessment of confidence through probability scores.

An early version of Probhap was also used with the Moleculo genome phasing technology introduced in:

V. Kuleshov et al., Whole-genome haplotyping using long reads and statistical methods, Nature biotechnology (2014) 32 (3), 261.

Installation

To install Probhap, clone the git repo and make sure the folder is in your PYTHONPATH:

git clone git@github.com:kuleshov/ProbHap.git;
cd probhap;
export PYTHONPATH="`pwd`:$PYTHONPATH"

Sample data

To download the data used for evaluating Probhap in the 2014 ECCB/Bioinformatics paper, type:

wget http://www.stanford.edu/~kuleshov/probhap/phasing-matrices.tar.gz
wget http://www.stanford.edu/~kuleshov/probhap/validation.tar.gz

tar -zxvf phasing-matrices.tar.gz
tar -zxvf validation.tar.gz

The first file contains phasing matrices derived from fosmid reads in the paper of Duitama et al. (2010). The second file contains the true phase of NA12878 determined by trio phasing.

Requirements

Probhap requires numpy and scipy-weave to run. During the first run, it will compile some inline C code using Weave. If these packages are not on your system, install them using pip or easy-install.

Running Probhap

Probhap takes as input a set of long reads in the same format as RefHap. See the paper by Duitama et al. (2010) and the documentation of the algorithm RefHap for more details.

To run Probhap, type:

python probhap.py \
  --reads data/input/chr22.matrix.SORTED \
  --parsed-reads chr22.probhap.reads \
  --phase chr22.probhap.out \
  --assignments chr22.probhap.assignments

This will take the reads in chr22.matrix.SORTED and produce phased blocks chr22.probhap.out in the same format as that of RefHap as well as chr22.probhap.assignments, which assigns reads to blocks. Before performing phasing, ProbHap does a merging and filtering step on the reads; the filtered reads are saved in chr22.probhap.reads.

Kuleshov (2014) also describes a post-processing heurstic on the phased blocks that typically results in a noticeable improvement in performance. To apply this post-processing step, type:

python probhap-postprocess.py \
  --filtered-reads chr22.probhap.reads \
  --assignments chr22.probhap.assignments
  --blocks chr22.probhap.out \
  --corrected-blocks chr22.probhap.corrected.out \

Reproducing experiments

To reproduce the experiments of Kuleshov (2014), use the script in the /experiments folder. This requires first downloading the RefHap and MixSIH packages and setting their paths in the Makefile. Also, make sure that you downloaded the sample data to /data.

To run the experiments, type:

make run
make other

Feedback

Please send comments and bug reports to Volodymyr Kuleshov.

nh13 / ProbHap