arcasHLA: high resolution HLA typing from RNA seq

arcasHLA performs high resolution genotyping for HLA class I and class II genes from RNA sequencing, supporting both paired and single-end samples.

Dependencies

arcasHLA requires the following utilities:

• Git Large File Storage
• coreutils

Make sure the following programs are in your PATH:

• Samtools v1.19
• bedtools v2.27.1
• pigz v2.3.1
• Kallisto v0.44.0
• Python 3.6

arcasHLA requires the following Python modules:

• Biopython v1.77 (or lower)
• NumPy
• SciPy
• Pandas

Test

In order to test arcasHLA partial typing, we need to roll back the reference to an earlier version. First, fetch IMGT/HLA database version 3.24.0:

./arcasHLA reference --version 3.24.0

Extract reads:

./arcasHLA extract test/test.bam -o test/output -t 8 -v

Genotyping (no partial alleles):

./arcasHLA genotype test/output/test.extracted.1.fq.gz test/output/test.extracted.2.fq.gz -g A,B,C,DPB1,DQB1,DQA1,DRB1 -o test/output -t 8 -v

Expected output in test/output/test.genotype.json:

{"A": ["A*01:01:01", "A*03:01:01"], 
 "B": ["B*39:01:01", "B*07:02:01"], 
 "C": ["C*08:01:01", "C*01:02:01"], 
 "DPB1": ["DPB1*14:01:01", "DPB1*02:01:02"], 
 "DQA1": ["DQA1*02:01:01", "DQA1*05:03"], 
 "DQB1": ["DQB1*02:02:01", "DQB1*06:09:01"], 
 "DRB1": ["DRB1*10:01:01", "DRB1*14:02:01"]}

Partial typing:

./arcasHLA partial test/output/test.extracted.1.fq.gz test/output/test.extracted.2.fq.gz -g A,B,C,DPB1,DQB1,DQA1,DRB1 -G test/output/test.genotype.json -o test/output -t 8 -v

Expected output in test/output/test.partial_genotype.json:

{"A": ["A*01:01:01", "A*03:01:01"], 
 "B": ["B*07:02:01", "B*39:39:01"],
 "C": ["C*08:01:01", "C*01:02:01"], 
 "DPB1": ["DPB1*14:01:01", "DPB1*02:01:02"], 
 "DQA1": ["DQA1*02:01:01", "DQA1*05:03"],
 "DQB1": ["DQB1*06:04:01", "DQB1*02:02:01"],
 "DRB1": ["DRB1*03:02:01", "DRB1*14:02:01"]}

Remember to update the HLA reference using the following command.

./arcasHLA reference --update

Usage

To see the list of available tools, simply enter arcasHLA. To view the required and optional arguments for any of the tools enter arcasHLA [command] -h.

• extract : Extracts reads mapped to chromosome 6 and any HLA decoys or chromosome 6 alternates.
• genotype : Genotypes HLA alleles from extracted reads (no partial alleles).
• partial : Genotypes partial HLA alleles from extracted reads and output from genotype (optional).
• reference : Update, specify version or force rebuilding of HLA reference.
• merge : merge genotyping output for multiple samples into a single json file.

Extract reads

arcasHLA takes sorted BAM files and extracts chromosome 6 reads and related HLA sequences. If the BAM file is not indexed, this tool will run samtools index before extracting reads. By default, extract outputs a single FASTQ file; use the --paired flag for paired-end samples.

arcasHLA extract [options] /path/to/sample.bam 

Output: sample.1.fq.gz, sample.2.fq.gz

Options:

• --single : single-end reads (default: False)
• --unmapped : include unmapped reads, recommended if the aligner used marks multimapping reads as unmapped (default: False)
• --log FILE : log file for run summary (default: sample.extract.log)
• --o, --outdir DIR : output directory (default: .)
• --temp DIR : temp directory (default: /tmp)
• --keep_files : keep intermediate files (default: False)
• -t, --threads INT : number of threads (default: 1)
• -v, --verbose : verbosity (default: False)

Genotype

From FASTQs

To predict the most likely genotype (no partial alleles), input the FASTQs produced by extract or the original FASTQs with all reads (experimental - use with caution).

arcasHLA genotype [options] /path/to/sample.1.fq.gz /path/to/sample.2.fq.gz

Output: sample.alignment.p, sample.em.json, sample.genotype.json

From intermediate alignment file

If you have previously run genotype on a sample, you can run genotype again directly from sample.alignment.p to retype without aligning with Kallisto again. This is useful if you want to try different populations, genes and other parameters.

arcasHLA genotype [options] /path/to/sample.alignment.p

Example .genotype.json

{'A': ['A*01:01:01', 'A*29:02:01'],
 'B': ['B*08:01:01', 'B*44:03:01'],
 'C': ['C*07:01:01', 'C*16:01:01'],
 'DQA1': ['DQA1*02:01:01', 'DQA1*05:01:01'],
 'DQB1': ['DQB1*02:01:01', 'DQB1*02:02:01'],
 'DRB1': ['DRB1*03:01:01', 'DRB1*07:01:01']}

Options

• -g, --genes GENES : comma separated list of HLA genes (ex. A,B,C,DQA1,DQB1,DRB1)
• -p, --population POPULATION : sample population, options are asian_pacific_islander, black, caucasian, hispanic, native_american and prior (default: Prior)
• --min_count INT : minimum gene read count required for genotyping (default: 75)
• --tolerance FLOAT : convergence tolerance for transcript quantification (default: 10e-7)
• --max_iterations INT : maximmum number of iterations for transcript quantification (default: 1000)
• --drop_iterations INT : number of iterations before dropping low support alleles, a lower number of iterations is recommended for single-end and low read count samples (default: paired - 10, single - 4)
• --drop_threshold FLOAT : proportion of maximum abundance an allele needs to not be dropped (default: 0.1)
• --zygosity_threshold FLOAT : threshold for ratio of minor to major allele nonshared count to determine zygosity (default: 0.15)
• --log FILE : log file for run summary (default: sample.genotype.log)
• --o, --outdir DIR : output directory (default: .)
• --temp DIR : temp directory (default: /tmp)
• --keep_files : keep intermediate files (default: False)
• -t, --threads INT : number of threads (default: 1)
• -v, --verbose : verbosity (default: False)
• --single : Include flag to indicate if single-end FASTQs (paired-end if missing)
• -l, --avg : Estimated average fragment length for single-end reads (default: 200)
• -s, --std : Estimated standard deviation of fragment length (default: 20)

Genotype - partial (optional)

Following genotyping, partial alleles can be predicted. This requires aligning the reads to an alternate, partial allele reference. The sample.genotype.json file from the previous step is required.

arcasHLA partial [options] -G /path/to/sample.genotype.json /path/to/sample.1.fq.gz /path/to/sample.2.fq.gz

Output: sample.partial_alignment.p, sample.partial_genotype.json

The options for partial typing are the same as genotype. Partial typing can be run from the intermediate alignment file.

Merge jsons

To make analysis easier, this command will merge all jsons produced by genotyping into a single table. All .genotype.json files will be merged into a single run.genotypes.tsv file and all .partial_genotype.json files will be merged into run.partial_genotypes.tsv. In addition, HLA locus read counts and relative abundance produced by alignment will be merged into a single tsv file.

arcasHLA merge [options]

Options

• --run RUN : run name
• --i, --indir DIR : input directory (default: .)
• --o, --outdir DIR : output directory (default: .)
• -v, --verbose : toggle verbosity

Convert HLA nomenclature

arcasHLA convert changes alleles in a tsv file from its input form to a specified grouped nomenclature (P-group or G-group) or a specified number of fields (i.e. 1, 2 or 3 fields in resolution). This file can be produced by arcasHLA merge or any tsv following the same structure:

subject	A1	A2	B1	B2	C1	C2
subject_name	A*01:01:01	A*01:01:01	B*07:02:01	B*07:02:01	C*04:01:01	C*04:01:01

P-group (alleles sharing the same amino acid sequence in the antigen-binding region) and G-group (alleles sharing the same base sequence in the antigen-binding region) can only be reduced to 1-field resolution as alleles with differing 2nd fields can be in the same group. By the same reasoning, P-group cannot be converted into G-group.

arcasHLA convert --resolution [resolution] genotypes.tsv

Options

• -r, --resolution RESOLUTION : output resolution (1, 2, 3) or grouping (g-group, p-group)
• -o, --outfile FILE : output file (default: ./run.resolution.tsv)
• -f, --force : force conversion for grouped alleles even if it results in loss of resolution
• -v, --verbose : toggle verbosity

Change reference

To update the reference to the latest IMGT/HLA version, run

arcasHLA reference --update

If you are running multiple tools to type HLAs, it can be helpful to use the same version of IMGT/HLA. You can select the version you like using the commithash from the IMGT/HLA Github.

arcasHLA reference --version [commithash]

If you suspect there is an issue with the reference files, rebuild the reference with the following command

arcasHLA reference --rebuild

Note: if your reference was built with arcasHLA version <= 0.1.1 and you wish to change your reference to versions >= 3.35.0, it may be necessary to remove the IMGTHLA folder due to the need for Git Large File Storage to properly download hla.dat.

rm -rf dat/IMGTHLA
arcasHLA reference --update

Options

• --update : update to latest IMGT/HLA version
• --version : checkout IMGT/HLA version using commithash
• --rebuild : rebuild HLA database
• -v, --verbose : verbosity (default: False)

Citations

Orenbuch R, Filip I, Comito D, et al (2019) arcasHLA: high resolution HLA typing from RNA seq. Bioinformatics doi:10.1093/bioinformatics/btz474

Orenbuch R, Filip I, Rabadan R (2020) HLA Typing from RNA Sequencing and Applications to Cancer. Methods Mol. Biol. doi: 10.1007/978-1-0716-0327-7_5 (https://link.springer.com/protocol/10.1007%2F978-1-0716-0327-7_5)