This repo describes assemblies produced by the Human Pangenome Reference Consortium from year 1 data. Assemblies for 47 samples are available. For information about data reuse and publicating with HPRC data please see the HPRC's Data Use Protocol.
All assemblies are accessioned at GenBank. They can be downloaded from public HPRC S3 bucket with no egress fee. Individual S3 and GCP URLs to each assembly can be found in the index file in this repo.
Assemblies are also available as a single 1.5GB file in the AGC format. Users need to download AGC to extract individual assemblies. For example:
# download all assemblies
curl -o HPRC-yr1.agc https://zenodo.org/record/5826274/files/HPRC-yr1.agc?download=1
# download precompiled AGC binary for Linux
curl -L https://github.com/refresh-bio/agc/releases/download/v1.1/agc-1.1_x64-linux.tar.gz|tar -zxvf - agc-1.1_x64-linux/agc
# list all samples
agc-1.1_x64-linux/agc listset HPRC-yr1.agc
# extract sample NA18906.1
agc-1.1_x64-linux/agc getset HPRC-yr1.agc NA18906.1 > NA18906.1.faFreeze 1 (v2) assemblies were uploaded to Genbank/NCBI and are now available. These assemblies are expected to be the final release of the year 1 assemblies and should be used for all analysis and pangenome work. A list of the current assemblies and their download links can be found in the index file assembly_index/Year1_assemblies_v2_genbank.index.
As part of the upload to Genbank, contaminated contigs are identified and dropped. Some contigs which are almost certainly contamination were not identified, however. In addition, leading/trailing hard-masked nucleotides were trimmed from the contigs -- resulting in a co-ordinate change for those contigs. Lastly, contigs were renamed by Genbank with their accession IDs. Files have been provided in the genbank_changes folder to document these changes.
Files in the genbank_changes folder:
- y1_genbank_contig_translation_table.txt: Translation table for contig names
- y1_genbank_dropped_contigs.txt: Contigs dropped by Genbank
- y1_genbank_remaining_potential_contamination.txt: Contigs that may be contamination
- y1_genbank_trimmed_sequence.txt: Sequence trimmed by Genbank
Assemblies were processed in AnVIL using publicly available workflows in the Human Pangenomics hpp_production_workflows repo
A summary of the process is below:
- Filter out sequences with adapters from HiFi data using HiFiAdapterFilt
- Run hifiasm v0.14
- For information about the input HiFi and parent Illumina data see the HPP Year 1 Data Freeze Repo
- Run assemblies through pipeline to identify MT sequences, adapter contamination, and viral/bacterial/fungal sequences
- Run by Kerstin Howe w/ a pipeline developed by Kerstin and James Torrance for the Tree of Life Programme at the Wellcome Sanger Institute
- Mask adapter sequences
- Use union of sequences found by decontamination pipeline and by mapping the PacBio SMRTBell dimer against the assemblies w/ minimap2
- Remove Mitochondrial and contaminated contigs from assemblies
- Add the "best" MT sequence back into the assembly
- best MT sequence defined by searching for the highest DP alignment score from minimap2 alignment against humand chrM reference
- Rename contigs to {sample_id}#{haplotype}#original_contig_id
- {haplotype} is 1 or 2 for paternal or maternal, respectively
After assembly masking, decontamination, and MT correction, assemblies underwent automated QC in AnVIL with the following tools:
Select metrics were extracted and placed into the automated_qc_results/ directory of this repo: raw values are in a CSV alongside charts for N50, hamming/switch error rates, QV values, and contig counts. Full results from automated QC are included next to the assemblies in both the AWS and GCP HPRC buckets.
For more information about the automated QC, please see the QC workflows in the HPP GitHub Repo.
- standard_qc
- standard_qc_no_qv
- run when no child Illumina data is available
Assemblies are available in the working directory of the HPRC S3 and GCP buckets and is organized by each trios' child sample ID:
── working/
└── HPRC/
└── HG00438/
└── raw_data/
└── assemblies/
└── year1_f1_assembly_v2_genbank/
└── HG00438.maternal.f1_assembly_v2_genbank.fa.gz
└── HG00438.paternal.f1_assembly_v2_genbank.fa.gz
└── year1_freeze_assembly_v2/
└── HG00438.maternal.f1_assembly_v2.fa.gz
└── HG00438.paternal.f1_assembly_v2.fa.gz
└── assembly_qc/
└── asmgene/
└── dipcall_v0.1/
└── dipcall_v0.2/
└── merqury/
└── quast/
└── yak/
Note that the current version of the assemblies is under year1_f1_assembly_v2_genbank/, but prior versions are also listed. Automated QC for the assemblies is under year1_freeze_assembly_v2/ since the QC was run on that version of the assemblies.
Raw hifiasm output (including GFAs) are included for the currently assembly release in each sample's working area under assemblies/hifiasm_v0.14_raw/.
A complete list of the paths to the assembly fastas in S3/GCP can be found in the assembly_index/ directory of this repo.
AnVIL users can access the data stored in GCP through the public AnVIL_HPRC workspace. Alternatively, data can be accessed directly from AWS or GCP, and data stored in the HPRC S3 Bucket can be accessed without egress fees.
- outstanding Screens run on the genbank version of the assemblies did not identify all contaminating contigs. A list of contigs that are likely contamination is provided in
genbank_changes/y1_genbank_remaining_potential_contamination.txt. Note that the assembly for HG02145 (paternal) has a large number of potential EBV contigs which have not been verified. - outstanding Three of the assemblies were artifically truncated due to an incomplete download from Genbank. The version of the assemblies on INSDCs is correct, but note that the version on S3/GCP is truncated. See below for details
- HG00733
- HG00733 Paternal size actual 3,042,264,782
- HG00733 Paternal size S3/GCP 3,041,978,016
- HG00733 Paternal number contigs actual 695
- HG00733 Paternal number contigs S3/GCP 681
- HG02630
- HG02630 Paternal size actual 3,053,402,297
- HG02630 Paternal size S3/GCP 3,053,354,263
- HG02630 Paternal number contigs actual 534
- HG02630 Paternal number contigs S3/GCP 532
- NA21309
- NA21309 Maternal size actual 3,036,775,048
- NA21309 Maternal size S3/GCP 3,036,587,447
- NA21309 Maternal number contigs actual 508
- NA21309 Maternal number contigs S3/GCP 501
- fixed: HG002 maternal contig h2tg000045l had a misjoin of chr1 and chr2 along with a large false duplication. Sample was reassembled using a new (v0.14.1) version of hifisasm to fix the misjoin issue. New assemblies released under v2.1.
- outstanding: HG01071 does not produce a MT contig
- manually fixed: HG02080 (maternal) had a misjoin in h2tg000053l
- h2tg000053l_1 : h2tg000053l [0, 41506502]
- h2tg000053l_2 : h2tg000053l [63683095, 92488067]
- outstanding: HG02080 (paternal) has a contig (h1tg000073l) where 4mb aligns to chr17 and 7mb aligns to chr19. It is unclear if this is a misjoin so it was not manually broken.
- manually fixed: The original paternal v2 assemblies for HG01123 & HG01358 each had a misjoin and a false duplication. This was fixed manually and released under v2.1
- Only these two samples (and HG002) have v2.1
- Note that the QC was run w/ the name v2 but v2.1 data was used for these two samples
- Below are the break points:
- HG01123: remove [94439457, contig end) of h1tg000013l
- HG01358: For h1tg000058l, keep [0, 95732608), remove [95732608, 150395342), keep [150395342, contig end)
* Jan 15, 2021: internal release of v1 assemblies
* Mar 08, 2021: internal release of v2 assemblies
* Added two new samples
* Added HiFiAdapterFilt preprocessing of HiFi data
* Switched to Hifiasm v0.14
* Added masking based on minimap2 alignments of SMRTBell adapter dimer
* Mar 18, 2021: fixed misjoin and false duplications in paternal assemblies for HG01123 & HG01358
* Apr 06, 2021: fixed misjoin and false duplication in HG002 maternal contig.
* Jun 23, 2021: internal release of v2 assemblies (Genbank version)
* Genbank identified and dropped around 3k contigs (almost all EBV)
* Assemblies were renamed to reflect their accesions in Genbank
* Some assemblies were trimmed to remove leading/trailing N's from adapter hard masking
