Output

The table

One row per candidate site, tab separated.

Column	Description
chrom, start, end	Host region covered by the anchors (0-based)
left_junction, right_junction	Junction estimated from forward and reverse anchors
left_breakpoint, right_breakpoint	Exact junction from split reads, where there are any
left_split_reads, right_split_reads	Reads clipped at that exact base
n_pairs, n_forward, n_reverse	Supporting pairs, split by anchor strand
top_taxid, top_taxid_fraction, n_taxa	Main foreign taxon and how consistent the calls are
mean_mapq	Mean mapping quality of the anchors
mean_mate_complexity	Mean 3-mer entropy of the unmapped mates (0 to 1)
taxon_anchored_pairs	Anchored pairs for this taxon across the genome
taxon_unanchored_pairs	Pairs for this taxon where neither mate mapped
expected_pairs, pvalue, qvalue	Enrichment test
expected_other_pairs, other_pvalue	Pairs from other taxa expected by chance, and the test for an excess of them
flags, filter	Flags raised and the overall filter result

A site passes when filter is PASS. Otherwise filter lists the reasons it failed.

BED

With --bed, the sites that passed every filter are written as BED: one line per site, named by the main taxid (for example taxid_562), with the number of supporting pairs as the score, capped at 1000 as the format requires. This loads straight into IGV or a genome browser.

VCF

With --vcf, every candidate site is written as a VCF 4.2 insertion record, so the calls can go through the same tools as other structural variants.

• POS is the last host base before the insertion. It comes from the split reads where there are any, otherwise from the anchors.
• ALT is the symbolic allele <INS> and SVTYPE=INS. hgt-site does not read the reference sequence, so REF is N.
• QUAL is the q-value on the phred scale, so QUAL of 13 is a q-value of 0.05 and larger numbers mean stronger enrichment. It is capped at 1000. Sites that failed a filter still carry their score, so a reader can rank the failures.
• FILTER is PASS or the failing flags.
• INFO holds TAXID, NPAIRS, NFWD, NREV, QVAL, SITE_START, SITE_END, SPLIT_LEFT, SPLIT_RIGHT, and the ONE_SIDED flag.
• A junction estimated from read pairs rather than found in split reads carries IMPRECISE, with CIPOS and CIEND giving how far the real junction could be from POS. The interval runs between the estimates from the two sides of the junction, or across the whole region the anchors cover when only one side has any. A junction found by split reads is exact and carries no interval.
• There is one sample column, named after the alignment file or after --sample-name, holding GT, PR and SR. PR is the supporting read pairs and SR the split reads crossing the junction. GT is always ./., because hgt-site counts the reads supporting a junction and does not work out how many copies of the host carry it.
• The header records the exact command and every threshold used, as ##hgt_siteCommand and ##hgt_siteSetting lines, so a file can be traced back to how it was made. Contig lengths come from the alignment header.

To keep only the sites that passed, or to filter on the score:

bcftools view -f PASS sample.hgt-site.vcf
bcftools view -i 'QUAL>50' sample.hgt-site.vcf

If the name given to --vcf ends in .gz, the file is block compressed and indexed with tabix, so it can be queried by region and used directly in a pipeline:

hgt-site call --bam sample.bam --kraken classified.txt --host-taxid 9606 \
    -o sample.hgt-site.tsv --vcf sample.hgt-site.vcf.gz
bcftools view sample.hgt-site.vcf.gz chr8:128000000-129000000

MultiQC

With --multiqc sample_mqc.json, hgt-site writes a custom-content table that MultiQC finds on its own, as long as the file name ends in _mqc.json. It holds the number of candidate sites, the number that passed, how many have an exact junction, and the read counts behind them.

Flags

Flag	Meaning
LOW_SUPPORT	Fewer than 3 supporting pairs
MIXED_TAXA	Less than 80% of pairs share one taxon, and the site holds more pairs from other taxa than background chimeras would explain
LOW_MAPQ	Mean mapping quality of the anchors below 20
LOW_COMPLEXITY	Unmapped mates have low 3-mer entropy (repeats or low-complexity sequence)
NOT_ENRICHED	Q-value above 0.05
ONE_SIDED	Anchors from only one side of the junction. Information only, and does not fail a site, since a long insertion can leave only one junction within insert range.