What this section is
One page per virus family, and eventually one per virus, recording what a virus is, what it does to people, and what exists against it. This build is the taxonomy pass: classification and host evidence only.
Searched, not searched, and none found are three different things. Wherever this section has not yet looked for something, it says “not searched yet”. It will say “searched — none found”, with the date, only once somebody has actually looked. A blank is never used to mean either.
Taxonomy
Every rank, the ICTV species binomial, the ICTV identifier and the exemplar isolate’s GenBank accession are copied by script from the ICTV Virus Metadata Resource, release MSL41.v1 / VMR_MSL41.v1.20260729.xlsx (released 29 July 2026, 17,554 species). Taxonomy is never hand-edited: a correction means re-running the fetch against a newer release.
ICTV renames species faster than common usage changes, so each virus carries its former and lay names as synonyms and the search box matches all of them. Three pages are named by hand because the ICTV exemplar name would mislead: the species holding SARS-CoV-2 is exemplified by the 2003 SARS-CoV, the dengue species by DENV-2, and the poliovirus species also contains coxsackie A viruses.
Baltimore class
Derived mechanically from the VMR’s own genome-composition field. Species whose field is ambiguous (a bare “ssRNA”) are left out rather than assigned a class by guesswork.
How a virus qualifies
Evidence of infecting humans or a named veterinary host.
The VMR records a host kingdom — “vertebrates”,
“plants”, “bacteria” — and never names a host
species, so human infection cannot be read off it. Instead, for each of the
4,295 vertebrate-host species, this build counts the GenBank nucleotide records
annotated /host="Homo sapiens", using the query template
txid{taxid}[Organism:exp] AND "Homo sapiens"[host]. A species qualifies at
2 records or more. Each virus
record stores its own count and query, so any entry can be re-checked against
NCBI without re-running anything.
This is a proxy, and it fails in both directions. Sequencing effort is not infection: a virus merely found in a human sample can clear the bar, and a genuine pathogen that nobody sequences can miss it. The figures here should be read as “what GenBank records say about provenance”, not as a definitive host range.
What is deliberately left out
160 species that clear the sequence test are excluded as non-pathogenic: Anelloviridae, Circoviridae, Endolinaviridae, Genomoviridae, Pecoviridae, Redondoviridae, Smacoviridae, and species named by the metagenomic-discovery convention (“Human associated cyclovirus 3”, “… sp.”). These are mostly commensal or environmental viruses found in human samples, typically with two records each, both from the paper that named them. Being found in a human sample is not the same as infecting humans, and including them would fill a fifth of the index with viruses no vaccine or antiviral will target.
This is an editorial judgement, not a finding. It is reversible in one command, and every excluded species is listed with its reason in the project's inclusion-review file.
Also absent, for now: viruses that infect only animals. Veterinary host screening is built and tested but switched off in this build, so their absence is a scope decision rather than evidence that none exist.
Vaccines
Vaccine entries are joined from this site’s own vaccine pipeline dataset, which remains the source of truth for a candidate’s stage. The stage is copied across, never re-derived, and the build fails if the two files ever disagree.
The pipeline records a target disease; this section needs a virus. No source states that mapping, so it is hand-maintained, and every decision is recorded — including the 47 diseases mapped to “no viral cause” (pneumococcal, meningococcal, the cancers) and the handful left unmapped because the label does not say which virus is meant. A candidate that maps to nothing appears in the build log rather than vanishing.
Two things the pipeline records are deliberately dropped rather than translated: a jurisdiction status of “phase 3” is a development stage, not a regulatory decision, and a free-text platform such as “Viral vector” does not say whether it replicates. Where a platform could not be mapped without guessing, the original wording is kept in the record’s notes and the field is left unset.
Antivirals and antibodies
Drugs come from the openFDA drug label endpoint, selected by the FDA’s own Established Pharmacologic Class. Those class names state the virus (“Hepatitis C Virus NS5A Inhibitor”), so the drug-to-virus link is the regulator’s classification rather than our reading of a label. Two generic classes — “Herpesvirus Nucleoside Analog DNA Polymerase Inhibitor” and “Nucleoside Analog Antiviral” — name no virus and are deliberately left unmapped.
Each indication carries the sentence from INDICATIONS AND USAGE that names the virus. This is US-only. A current US label means the product is approved in the United States; a drug approved only elsewhere will not appear here, and its absence is not evidence that it does not exist.
Impact figures
Every figure is the one its source printed, with the sentence it came from. Nothing is pooled, averaged, converted or recomputed, and no figure is written from memory. The source page is cached, and the build re-reads that cache and checks the quote appears in it verbatim; an estimate whose quote does not match is held back and never reaches a page. Changing a single digit is enough to fail that check.
Charts. Where a virus has a disease-risk page, its “What it does” section draws that page's records: a strip of vaccine-effectiveness estimates on a 0–100 per cent axis, one mark per study at the figure it reported with the interval it reported, coloured by study design; headline burden figures with the sentence they came from; and, on the index, a shared axis of reproduction numbers and incubation periods. The charts are inline SVG generated at build time from the same records as the tables. Nothing is pooled or averaged: the lowest and highest figures are named as two studies, not a range of truth, and a hollow mark means the source gave no interval. A percentage of “cases” and a percentage of “infections” are kept apart by colour, because the first is always larger. The table under each chart is the accessible view of the same data.
Animal viruses and their vaccines
The veterinary section is built the same way, screening each ICTV species against fourteen farmed and companion host species instead of against people. Its vaccines are the current UK marketing authorisations published by the Veterinary Medicines Directorate, taken from that register's own bulk download. UK only: a vaccine licensed elsewhere and not in the UK does not appear.
The two sections are kept apart on purpose. Every indication records who the product is authorised for; the schema requires it, the validator rejects a veterinary product that cites a human source or the reverse, and no page renders a product from the other section. The reason is that the harm is asymmetric — an animal vaccine read as a human one could lead someone to believe protection exists for a person where it does not. A virus that infects both appears in both sections, with each page showing only its own products.
Screening animal hosts also corrected the human index. Sixteen viruses had been included on a handful of human sequence records and turned out to be animal viruses with incidental human reads — African swine fever had 2 human records against 5,089 from pigs. Their human pages now redirect to the veterinary ones.
Occurrence maps
Six diseases carry a map: measles, rubella, mumps, polio, yellow fever and Japanese encephalitis. The data is the WHO/UNICEF reported-case series this site already publishes on its country pages, shaded by the cases each national authority reported for the latest year.
What these maps are not. They are not modelled distribution, and they are deliberately not a map of where a virus has been sequenced — sequencing tracks laboratory funding, so shading it geographically would make well-funded countries look like hotspots. Reported cases undercount true infections, and a zero can mean elimination or simply that nobody reported; the data-quality grade on each country series says which. A virus with no WHO reported-case series gets no map rather than an empty one.
Not in this build
Transmission routes, reservoirs, vectors and geography; candidate antivirals from trial registries; approvals outside the US; and impact figures for most viruses. Each section says so on the page rather than showing an empty cell.
Sources and licences
- ICTV Virus Metadata Resource (MSL41.v1 / VMR_MSL41.v1.20260729.xlsx) — taxonomy. Freely reusable with attribution.
- NCBI Taxonomy and GenBank — taxon identifiers, synonyms and host provenance counts. US Government work, public domain.
This page and the rest of the section are CC BY 4.0. Extract dates: ICTV 2026-10-30, NCBI 2026-10-30.