HapDup (haplotype duplicator) is a pipeline to convert a haploid long read assembly into a diploid assembly. HapDup works with either Oxford Nanopore or PacBio HiFi. The reconstructed haplotypes preserve heterozygous structural variants (in addition to small variants) and are locally phased.

HapDup takes as input a haploid long-read assembly, such as produced with Flye or Shasta. Currenty, ONT reads (Guppy 5+ recommended) and PacBio HiFi reads are supported.

HapDup is currently designed for low-heterozygosity genomes (such as human). The expectation is that the assembly has most of the diploid genome collapsed into a single haplotype. For assemblies with partially resolved haplotypes, alternative alleles could be removed prior to running the pipeline using purge_dups. We expect to add a better support of highly heterozygous genomes in the future.

The first stage is to realign the original long reads on the assembly using minimap2. We recommend to use the latest minimap2 release.

minimap2 -ax map-ont -t 30 assembly.fasta reads.fastq | samtools sort -@ 4 -m 4G > lr_mapping.bam
samtools index -@ 4 assembly_lr_mapping.bam

HapDup is available on the Docker Hub.

If Docker is not installed in your system, you need to set it up first following this guide.

Next steps assume that your assembly.fasta and lr_mapping.bam are in the same directory, which will also be used for HapDup output. If it is not the case, you might need to bind additional directories using the Docker's -v / --volume argument. The number of threads (-t argument) should be adjusted according to the available resources. For PacBio HiFi input, use --rtype hifi instead of --rtype ont.

cd directory_with_assembly_and_alignment
HD_DIR=`pwd`
docker run -v $HD_DIR:$HD_DIR -u `id -u`:`id -g` mkolmogo/hapdup:0.12 \
  hapdup --assembly $HD_DIR/assembly.fasta --bam $HD_DIR/lr_mapping.bam --out-dir $HD_DIR/hapdup -t 64 --rtype ont

Alternatively, you can use Singularity. First, you will need install the client as descibed in the manual. One way to do it is through conda:

conda install singularity

Next steps assume that your assembly.fasta and lr_mapping.bam are in the same directory, which will also be used for HapDup output. If it is not the case, you might need to bind additional directories using the --bind argument. The number of threads (-t argument) should be adjusted according to the available resources. For PacBio HiFi input, use --rtype hifi instead of --rtype ont.

singularity pull docker://mkolmogo/hapdup:0.12
HD_DIR=`pwd`
singularity exec --bind $HD_DIR hapdup_0.12.sif \
  hapdup --assembly $HD_DIR/assembly.fasta --bam $HD_DIR/lr_mapping.bam --out-dir $HD_DIR/hapdup -t 64 --rtype ont

The output directory will contain:

• hapdup_dual_{1,2}.fasta - dual assembly
• phased_blocks_hp{1,2}.bed - phased blocks coordinates (in dual assmeblies)
• hapdup_phased_{1,2}.fasta - haplotype-resolved assmebly

Becuase the pipeline is only using long-read (ONT/HiFi) data, it typically does not achieve chromosome-level phasing, as opposed to the approaches that use Hi-C or Strand-Seq. Hapdup represents the diploid assembly with in two modes: dual and phased.

Dual assembly will have the same contiguity as the original input, but may contain phase switches. Location of the switches are given in the bed files. Dual assembly haplotypes will contain homozogous and heterozygous varinats (small and structural). Therefore this type of output is convenient for variant calling. Read more about this assembly representation here.

Alternatively, hapdup_phased_* files contain haplotype-resolved contigs. Phases are assigned arbitrary and different contigs in one file may not be in synch. Because phasing N50 is typically lower than haploid assembly N50, this representation will require original contigs to be broken and is therefore more fragmented.

1. HapDup starts with filtering alignments that are likely originating from the unassembled parts of the genome. Such alignments may later create false haplotypes if not removed (e.g. if reads from a segmental duplication with two copies can create four haplotypes).

2. Afterwards, PEPPER is used to call SNPs from the filtered alignment file

3. Then we use Margin to phase SNPs and haplotype reads

4. We then use Flye to polish the initiall assembly with the reads from each of the two haplotypes independently

5. Finally, we find (heterozygous) breakpoints in long-read alignments and apply the corresponding structural changes to the corresponding polished haplotypes. Currently, it allows to recover large heterozygous inversions.

We evaluated HapDup haplotypes in terms of reconstructed structural variants signatures (heterozygous & homozygous) using the HG002 for which the curated set of SVs is available. We used the recent ONT data basecalled with Guppy 5.

Given HapDup haplotypes, we called SV using hapdiff. We also compare SV set against hifiasm assemblies, even though they were produced from HiFi, rather than ONT reads. Evaluated using truvari with -r 2000 option. GT refers to genotype-considered benchmarks.

Yak k-mer based evaluations:

Given a minimap2 alignment, Hapdup runs in ~400 CPUh and uses ~80 Gb of RAM.

If you prefer, you can install from source as follows:

#create a new conda environemnt and activate it
conda create -n hapdup python=3.8
conda activate hapdup

#get HapDup source
git clone https://github.com/fenderglass/hapdup
cd hapdup
git submodule update --init --recursive

#build and install Flye
pushd submodules/Flye/ && python setup.py install && popd

#build and install Margin
pushd submodules/margin/ && mkdir build && cd build && cmake .. && make && cp ./margin $CONDA_PREFIX/bin/ && popd

#build and install PEPPER and its dependencies
pushd submodules/pepper/ && python -m pip install . && popd

To run, ensure that the conda environemnt is activated and then execute:

conda activate hapdup
./hapdup.py --assembly assembly.fasta --bam lr_mapping.bam --out-dir hapdup -t 64 --rtype ont

The major parts of the HapDup pipeline are:

• PEPPER
• Margin
• Flye

The pipeline was originally developed at Paten lab at UC Santa Cruz. The work continues at Kolmogorov lab at NCI.

Code contributors:

• Mikhail Kolmogorov

PEPPER/Margin/Shasta support:

• Kishwar Shafin
• Trevor Pesout
• Paolo Carnevali

If you use HapDup in your research, the most relevant papers to cite are:

• Kolmogorov, Billingsley et al, "Scalable Nanopore sequencing of human genomes provides a comprehensive view of haplotype-resolved variation and methylation". bioRxiv (2023) doi:10.1101/2023.01.12.523790

• Kishwar Shafin, Trevor Pesout, Pi-Chuan Chang, Maria Nattestad, Alexey Kolesnikov, Sidharth Goel, Gunjan Baid et al. "Haplotype-aware variant calling enables high accuracy in nanopore long-reads using deep neural networks." bioRxiv (2021). doi:10.1101/2021.03.04.433952

• Mikhail Kolmogorov, Jeffrey Yuan, Yu Lin and Pavel Pevzner, "Assembly of Long Error-Prone Reads Using Repeat Graphs", Nature Biotechnology, 2019 doi:10.1038/s41587-019-0072-8

HapDup is distributed under a BSD license. See the LICENSE file for details. Other software included in this discrubution is released under either MIT or BSD licenses.

A preferred way report any problems or ask questions is the issue tracker.

In case you prefer personal communication, please contact Mikhail at mikolmogorov@gmail.com.