For general details including installation of QUILT, citation, versions, and changelog, see the main QUILT README.
- Introduction
- Installation
- Quick start run
- Input and output formats
- Help, options and parameters
- Reference packages
- Preparing a reference package
QUILT-HLA is program for rapid HLA imputation from low-coverage sequence. QUILT-HLA uses information from reads within an HLA locus through direct read mapping, and uses information from imputation using a labelled haplotype reference panel using reads outside an HLA locus. QUILT-HLA is highly accurate, including at low coverage, and has been tested with coverages as low as 0.1X, with higher coverage slightly increasing accuracy through more reads being available to inform direct read mapping. Imputation results are most accurate for samples most closely matching the reference panel, but are generally high across populations. Imputation results using low-coverage sequence and QUILT-HLA generally outperform imputation from genotyping microarray input. Further details and detailed evaluations are available in the QUILT paper.
QUILT-HLA is installed through the installation of QUILT. No additional installation is necessary.
A quick start to ensure QUILT-HLA is properly installed and working can be performed using the following.
First, download some prepared reference panel data. This reference panel package was prepared as described in Preparing a reference package, and uses data from IPD-IGMT version 3.39, 1000 Genomes Project haplotypes (20201028), and 1000 Genomes Project HLA types (20181129).
This example uses a reference panel data package with some of the reference samples excluded(e.g. NA12878). For normal use, use a panel without samples removed. See Reference packages for more details.
wget http://www.stats.ox.ac.uk/~rdavies/QUILT_HLA_reference_package_samples_excluded_2021_04_08.tar ## or curl -O
tar -xvf QUILT_HLA_reference_package_samples_excluded_2021_04_08.tar
Download some example bams
wget http://www.stats.ox.ac.uk/~rdavies/QUILT_HLA_example_bams_2021_04_08.tar
tar -xvf QUILT_HLA_example_bams_2021_04_08.tar
HLA imputation for a particular region (here A) can be done as follows
HLA_GENE="A"
./QUILT_HLA.R \
--outputdir=quilt_output \
--bamlist=bamlist.txt \
--region=${HLA_GENE} \
--prepared_hla_reference_dir=quilt_hla_reference_panel_files_2021_04_08 \
--quilt_hla_haplotype_panelfile=quilt_hla_reference_panel_files_2021_04_08/quilt.hrc.hla.${HLA_GENE}.haplotypes.RData \
--hla_gene_region_file=hla_ancillary_files/hlagenes.txt \
--dict_file=hla_ancillary_files/GRCh38_full_analysis_set_plus_decoy_hla.dict
The above imputes three samples, with true HLA types as follows
Population Sample.ID HLA.A.1 HLA.A.2
CEU NA12878 01:01 11:01
ASW NA19625 02:01 23:17
ASW NA19700 03:01 30:01
The output of QUILT-HLA will vary slightly run to run, as it uses random sampling, but should look approximately like the following (from file quilt_output/quilt.hla.output.combined.all.txt)
sample_number sample_name bestallele1 bestallele2 post_prob sums
1 NA12878 A*01:01 A*11:01 0.999996657444772 0.999996657444772
2 NA19625 A*02:01 A*23:01 0.673515530204319 0.673515530204319
2 NA19625 A*02:01 A*23:17 0.326258885270228 0.999774415474547
3 NA19700 A*03:01 A*30:01 0.999999942895535 0.999999942895535
Here we see that NA12878 is correctly imputed, as is NA19700, while for NA19625, the first allele is imputed correctly (A*02:01), while the second one is imputed less confidently, with the most confident allele (A*23:01, confidence 0.67) is incorrect, while the second allele (A*23:17, confidence 0.3262) is correct. More details about the output formats are given in Output.
- Bams. Given as a bamlist (i.e. a file with one row per sample, the path to the bam)
- Reference package and ancillary files. For available pre-made options, see Reference packages. For more detail about building your own, see Preparing haplotype files. For help in making a reference package, or any other questions or bugs, please feel free to email or file an issue on github.
Output is given as text files, with default names quilt.hla.output.<combined/onlystates>.<all/topresult>.txt. Here we begin by revisiting quilt.hla.output.combined.all.txt, explaining the output, and then explain the variations in the other files
sample_number sample_name bestallele1 bestallele2 lhoods sums
1 NA12878 A*01:01 A*11:01 0.999996657444772 0.999996657444772
2 NA19625 A*02:01 A*23:01 0.673515530204319 0.673515530204319
2 NA19625 A*02:01 A*23:17 0.326258885270228 0.999774415474547
3 NA19700 A*03:01 A*30:01 0.999999942895535 0.999999942895535
The first two columns are straightforward, of sample_number, being the 1-based integer index of the sample from the original bamlist, and sample_name, being the sample name of that sample as taken from the BAM header. Next, we have the pair of imputed alleles, given in columns bestallele1 and bestallele2. Finally, post_prob gives the posterior probability of the combination of alleles, and sums gives the sum of posterior probabilities of successive alleles (i.e. cumulative sum of post_prob across pairs of alleles). Pairs of alleles are outputted until the sums argument exceeds the QUILT-HLA parameter summary_best_alleles_threshold with default value 0.99.
The difference between all/topresult is that topresult only outputs the single most likely pair of alleles for each sample, while all continues until the sums value exceeds the threshold as explained just above.
The difference between combined/onlystates is that combined uses information from all reads using both read mapping and imputation using a labelled haplotype reference panel, while onlystates uses only the imputation and not direct read mapping. As such you are recommended to use combined as your default file to use.
For a full list of options, query ?QUILT::QUILT_HLA, or alternatively, type
./QUILT_HLA.R --help
Reference packages built
Max N=5132 haplotyes built using QUILT 0.1.5, IPD-IGMT version 3.39, 1000 Genomes Project haplotypes (20201028), and 1000 Genomes Project HLA types (20181129).
http://www.stats.ox.ac.uk/~rdavies/QUILT_HLA_reference_package_2021_04_09.tar
An example of this is presented in detail in example/QUILT_hla_reference_panel_construction.Md, which was used to make the reference panel package from 1000 Genomes Project data presented above.