How hgt-site compares with other tools¶
Finding foreign DNA inserted into a host genome is an old problem with several existing tools, and some of them are better than hgt-site at the thing they were built for. This page is here so that someone can decide what to use, including deciding not to use hgt-site.
Dates were read from the repositories on 2 November 2026.
| Tool | What it looks for | Latest release | Last commit |
|---|---|---|---|
| LGTSeek | Bacterial lateral gene transfer in host sequencing data | v1.08, January 2015 | June 2016 |
| IGS lgtseek-pipeline | The same method, packaged as a pipeline | none tagged | February 2022 |
| VIRUSBreakend | Viral integration, using single breakends | v2.13.2, February 2022 | May 2025 |
| ViFi | Viral integration, using HMMs over viral families | none tagged | July 2021 |
| nf-core/hgtseq | A pipeline that aligns reads and classifies the unmapped ones | 1.1.0, May 2023 | actively developed |
| hgt-site | Bacterial or viral insertion sites, tested against chance chimeras | 0.3.0, November 2026 | actively developed |
When another tool is the better choice¶
You are looking for viral integration and want the best available evidence. Use VIRUSBreakend. It assembles the breakend itself rather than inferring a position from read pairs, it handles the viral taxonomy and the repetitive viral genomes that cause trouble, it is part of GRIDSS and so benefits from a mature structural variant caller, and it has been published and used on real cancer genomes. hgt-site has not been compared with it. Until that comparison exists, anyone who tells you hgt-site is better for viral integration, including this documentation, is guessing.
You want horizontal gene transfer between bacterial genomes over evolutionary time. None of the tools above address that. It is a phylogenetic question and needs phylogenetic methods.
You have long reads. A single read spanning a junction is stronger evidence than anything a paired-read method can produce.
You want a whole workflow rather than one step. A pipeline will align your reads, classify the unmapped ones, run quality control and assemble a report. hgt-site does one step of that and expects the inputs to exist already.
What hgt-site does that the others do not¶
It tests each candidate against the number of chance chimeras expected there. The tools above filter on read support, on alignment evidence, or on assembly of the junction. None of them model the rate at which library preparation joins a host fragment to a foreign fragment by chance, which is the dominant source of false sites. On the benchmark datasets, filtering on read support alone reported 24,861 sites of which 390 were real.
It works on any foreign taxon, not a curated reference set. VIRUSBreakend and ViFi both work against viral references. hgt-site takes whatever Kraken2 reported, so a bacterium nobody expected is treated the same as one that was.
It keeps the rejected candidates. Every candidate position stays in the output with the reason it failed, so a region can be checked without rerunning anything.
It is small and installs with pip. One dependency, no Java, no reference bundle, no workflow engine. For LGTSeek in particular, which has had no commit since 2016, getting it to run at all now takes real work.
What has actually been compared¶
Only LGTSeek, and only its read classification step rather than its later BLAST-based filtering. That comparison is in the benchmarks, with the scripts and raw results in the repository.
Comparing hgt-site with VIRUSBreakend on viral integration data is the obvious next benchmark and has not been done. It is a harder test than the LGTSeek comparison and may well go against hgt-site.
The honest summary¶
hgt-site is a focused tool with a specific statistical idea behind it, tested carefully on simulated data and not at all on real data. VIRUSBreakend is a mature, published tool for the viral case. If your problem is viral integration in a real sample today, use VIRUSBreakend. If your problem is bacterial sequence in a host sample, or you want to see the reasoning behind each call, hgt-site is worth half an hour.