Probabilistic HLA Typing

Paper: Prohlatype: A Probabilistic Framework for HLA Typing ¹

This project provides a set of tools to calculate the full posterior distribution of HLA types given read data.

Instead of:

	A1  	A2  	B1  	B2  	C1	    C2  	Reads	Objective
0	A*31:01	A*02:01	B*45:01	B*15:03	C*16:01	C*02:10	538.0	513.79

one can calculate:

Allele 1	Allele 2	Log P	P
A*02:05:01:01	A*30:114	-23046.81	0.5000
A*02:05:01:01	A*30:01:01	-23046.81	0.5000
A*02:05:01:01	A*30:106	-23103.15	0.0000
A*02:05:01:02	A*30:114	-23146.35	0.0000
...
B*07:36	B*57:03:01:02	-13717.33	0.5000
B*07:36	B*57:03:01:01	-13717.33	0.5000
B*07:36	B*57:03:03	-13804.74	0.0000
B*27:157	B*57:03:01:02	-13816.17	0.0000
...
C*06:103	C*18:10	-11936.35	0.3338
C*06:103	C*18:02	-11936.36	0.3331
C*06:103	C*18:01	-11936.36	0.3331
C*15:102	C*18:02	-11951.72	0.0000

How:

There are three options to obtain the software:

1. If you are running on Linux, standalone binaries are available with each release.

2. Use the linked Docker image.

3. Build the software from source:

   a. Install opam.

   b. Make sure that the opam packages are up to date:

    $ opam update
   

   c. Make sure that you're on the relevant compiler:

    $ opam switch 4.06.0
    $ eval `opam config env`
   

   d. Get source:

    $ git clone https://github.com/hammerlab/prohlatype.git prohlatype
    $ cd prohlatype
   

   e. Install the dependent packages:

    $ make setup
   

   f. Build the programs (afterwards they'll be in _build/default/src/apps):

    $ make
   

Make sure that you have IMGT/HLA available:

$ git clone https://github.com/ANHIG/IMGTHLA.git imgthla

"Prohla"-typing:

1. Create an imputed HLA reference sequence via align2fasta. This step makes sure that all alleles have sequence information that spans the entire locus. This way, reads that originate from a region for which we normally do not have sequence information will still align (in the next filtering step), albeit poorly:

    $ align2fasta path-to-imgthla/alignments -o imputed_hla_class_I
   

   This step needs to be performed only once, per each IMGT version. Run $align2fasta --help for further information.

2. Filter your data against the reference, by first aligning. Ex:

    $ bwa mem imputed_hla_class_I.fasta ${SAMPLE}.fastq | \
        samtools view -F 4 -bT imputed_hla_class_I.fasta -o ${SAMPLE}.bam
   

   While fundamentally, the algorithms here are alignment based. They're too slow to run for all sequences. Sequences that do not originate from the HLA-region would just act as background noice.

3. and then convert aligned reads back to FASTQ:

    $ samtools fastq ${SAMPLE}.bam > ${SAMPLE}_filtered.fastq
   

4. Infer types (see $ multi_par --help for further details):

    $ multi_par path-to-imgthla/aignments ${SAMPLE}_filtered.fastq -o ${SAMPLE}_output.tsv
   

Note: The script src/scripts/run-example-docker.sh provides an end-to-end example of the above. It depends only on docker, wget, and git; it fetches the data and runs everything in a docker container (see sh src/scripts/run-example-docker.sh help).

1: All versions of this software after 0.8.0 incorporate an important coverage likelihood that is not described in the previous paper. At the moment a short addendum describing the approach is in limbo, please contact me by email for a reference.