IoanFilip2
commented
3 years ago Thanks for your interest in our tool and for your constructive feedback today. Regarding your specific questions here:
- arcasHLA works well with all reference genome builds and annotations, and there isn't really a preference or suggested usage from our experience; in our manuscript, we make a point specifying that ALT sequences and HLA decoys are dealt with in a satisfactory manner if included in the genome annotation (namely, the ALT sequences are checked by default and those reads which map to such ALT sequences are not discarded from the extraction step for the purpose of HLA genotyping);
- it is likely better not to exclude any non-chromosome sequences or to alter the genomic reference in any way prior to read mapping; the mapping output from STAR (which is the input bam for arcasHLA) is dealt with accordingly at the extraction step; we also recommend using the --unmapped flag to increase the number of reads considered at the pseudo-alignment and genotyping steps.
How different are the genotypes called with arcasHLA between your runs? Is there a big discrepancy in every sample or in some samples only? Is there a big discrepancy in the first two fields or only in the third (or fourth) field? Are you genotyping class I, class II or both?
slowkow
chrisdphd
Could I please ask if you might be willing to discuss a few questions?
chr6in each BAM file, I think the answer is "yes.")I tried two different options.
Run 1: GENCODE
I downloaded this sequence:
ftp://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_37/GRCh38.p13.genome.fa.gz
Then, I mapped reads with STAR and ran arcasHLA on the BAMs.
Run 2: Filtered UCSC
I downloaded this sequence:
https://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/
First, I discarded all sequences from the fasta file with names that are not in the list
chr1-23,X,Y.Then, I mapped reads with STAR and ran arcasHLA on the BAMs.
Results
arcasHLA called a greater number of genotypes in the UCSC run than the GENCODE run. I don't know what the true genotypes might be. Many of the genotypes do not match between the two runs.
I think the main reason for the difference between the two runs is whether or not we include the "alternate" or "patch" or "scaffold" sequences. For the GENCODE run, the alternate sequences were included in the read-mapping step. For the UCSC run, the alternate sequences were not included.
I haven't tested to see if the chromosome sequences (
chr1-23,X,Y) are identical between GENCODE and UCSC, but I might guess that they are very similar or identical.