You can find Locityper genotype predictions and preprocessed data for the 1000 genomes project (1KGP) samples, as well as benchmarking data at Zenodo (opens in a new tab).

Locityper genotypes for 1KGP

Genotyping results can be found in the 1kgp_genotypes.csv.gz file with the following columns:

• sample, locus,
• genotype: Primary genotype prediction,
• quality: Phred-scale probability of a non-primary genotype (possibly very similar) being correct,
• total_reads: How many reads were recruited for genotyping this locus,
• unexpl_reads: How many reads were not explained by the primary genotype,
• weight_dist: Weighted distance between the primary genotype prediction and other genotypes. Here, distance was calculated as the Jaccard distance between multisets of (15,15)-minimizers. Distances are weighted according to the secondary genotype probabilities.
• warnings: Other heuristic warnings.

Genotypes were calculated using Locityper v0.14.4-0 based on 3202 Illumina WGS datasets.

Target loci database

A database of 327 target loci can be found by decompressing db_327.tar.gz. For each locus, the following files can be useful:

• ref.bed: reference coordinates of the locus,
• discarded_haplotypes.txt: haplotypes, identical to others and discarded from futher analysis,
• haplotypes.fa.gz: all unique HPRC haplotypes for the locus,
• haplotypes.paf.gz: pairwise alignments of all haplotypes to each other.

Background parameters for the full 1KGP cohort

Locityper-preprocessed background parameters for 3202 Illumina WGS datasets can be found in bg-1kgp.tar.gz. After decompressing, corresponding directories used during genotyping as locityper genotype ... -p bg-1kgp/SAMPLE.

Locityper benchmarking

Locityper benchmarking results can be found in the locityper-benchmarking.tar.gz archive. Except for the trio comparison, benchmarking was performed based on the HPRC samples using Locityper v0.14.2-4.

QV subdirectory contains 9 files, of them 4 full_* files with Locityper evaluation using full reference panel, loo_* files for Locityper in the leave-one-out (LOO) configuration, and nygc.csv.gz for evaluating NYGC 1KGP call set haplotypes.

Each file has three lines for each sample and locus: based on the query_type column, two hap lines represent haplotype evaluation and gt line shows diploid genotype evaluation. In addition to previously explained columns, the files may have following columns:

• edit, size: edit distance and alignment size in the haplotype/genotype alignment,
• div = edit / size,
• qv: Phred-like transformation of div,
• avail_*: in the LOO-setting, what is the similarity between the actual and the closest available haplotypes/genotypes in the LOO database.

Trio concordance

Trio concordance based on the 1KGP samples can be found in the trio_conc.csv.gz file. For each child (indiv column) and each locus the file shows concordance between child-parent haplotypes/genotypes. Last (expl) column explains which child-parent haplotype pairs were matched together.

Benchmarking HLA/KIR calling

HLA/KIR results can be found in the HLA subfolder. immuannot.csv.gz contains baseline HLA/KIR annotation for the HPRC samples. alleles1 and alleles2 represent alleles on the first and second haplotype of the corresponding samples. If a haplotype contains a novel allele, it can be annotated with several close alleles, combined via |.

Three other files evaluate HLA/KIR calling accuracy, where each sample and locus can be written several times, according to the number of haplotypes. For Locityper, column qual is either zero or one: latter if there are no warnings, weighted distance is at most 30, and there are less than 1000 or less than 20% unexplained reads.

Accuracy column shows how many allele fields were matched, while base_length shows how many fields there are in the baseline allele. Finally, avail column shows if this the first two baseline fields are available in the LOO setting.

Variant calling comparison

Variant calling comparison between Locityper, 1KGP and Pangenie can be found in the varcall-eval.csv.gz file. There, number of true/false positive/negative variant calls are shown for each combination of locus,tool,sample, variant type and quality threshold (0/1/10).