# NCBI Viral Genotyping Tool: How to Choose The NCBI Viral Genotyping Tool is a free web server that assigns a genotype or subtype to a viral [nucleotide sequence](/knowledge/molecular-biology/nucleotide-sequence) by comparing it with BLAST against a curated panel of reference sequences, then painting the match along your query as a color-coded map that exposes recombinant mosaics [1]. The rule of thumb for choosing: treat the NCBI tool as your fast first pass for HIV-1, HCV, and HBV sequences, use it with custom references for anything else, and confirm any result that will end up in a paper, a treatment decision, or an outbreak report with a phylogenetic tree you build and inspect yourself. This guide walks through the NCBI viral genotyping tool options in detail, compares them head to head with the rival rapid subtyping servers, and gives you concrete decision criteria and worked scenarios so the choice becomes mechanical rather than a matter of habit. ## What the NCBI Viral Genotyping Tool Actually Does The tool was published in 2004 in [Nucleic Acids](/blog/guides/nucleic-acids) Research as a web-based genotyping resource built around a simple and durable idea. You submit a query nucleotide sequence. The server uses BLAST to compare that query against reference sequences for known genotypes. The query is then broken into segments, and each segment is classified against the reference panel. Segmenting is the design feature that separates this tool from a plain BLAST lookup. A recombinant virus, meaning a genome stitched together from lineages that originally belonged to different genotypes, will show one genotype color across part of its length and a different color across another part. The results are displayed graphically with color-coded genotypes, so the genotype of any portion of the query can be read off directly [1]. Out of the box, the tool ships with predefined reference genotype panels for three major viral pathogens: HIV-1, HCV (hepatitis C virus), and HBV (hepatitis B virus). You can also supply your own reference sequences and run them alongside or instead of the built-in panels [1]. The original form page lived at the NCBI projects path given in the publication, and as with any NCBI resource you should locate the current entry point by searching the NCBI site for the genotyping tool if the historical link has moved. It helps to place this tool inside the wider NCBI ecosystem, because the choice you are making is really a choice among layers of the same database infrastructure. GenBank stores the raw sequence records. BLAST compares your query to everything in a chosen database. The non-redundant nucleotide database (nr) is what HCV reference-method studies have actually used for genotype calls, alongside dedicated callers [2], [3]. The NCBI Translate tool converts your nucleotide sequence into [protein sequence](/blog/guides/protein-sequence), which is a routine early step in HIV envelope workflows before downstream analysis [4]. The genotyping tool sits on top of this stack as a domain-specific layer that already knows what the reference genotypes are called. ## Your NCBI Viral Genotyping Tool Options at a Glance There are four practical ways to get a viral genotype call from sequence data, and the first two are variations of the same tool. This table is the core of the decision. | Option | What it does | Best for | Main drawback | |--|--|--|--| | NCBI tool, predefined references | BLASTs your query against built-in HIV-1, HCV, or HBV genotype panels and segments the query to reveal recombination [1] | Fast genotype and recombination scans on the three supported viruses | No predefined panels beyond HIV-1, HCV, and HBV, and imperfect accuracy on non-B HIV clades [5] | | NCBI tool, custom references | Runs the same segmentation logic against reference sequences you supply [1] | Regional variants, non-human viruses, any virus without a built-in panel | Reference set quality and labeling are entirely your responsibility | | Other rapid web tools (Stanford, REGA, Geno2pheno, COMET, HBV-specific servers) | Rule-based, phylogeny-in-tree, or alignment-score-based calls on curated databases [5], [3], [6] | Cross-checking, virus-specific subtyping depth | Every tool has a different blind spot, and agreement between two tools does not guarantee correctness [7] | | Phylogenetic analysis you run yourself | Builds a tree with your query plus reference sequences and reads the placement | Recombinant forms, unusual variants, publication-grade calls | Slower, requires alignment and tree-building competence | The short version of the choice: if your virus is HIV-1, HCV, or HBV, start with the NCBI tool's predefined panel and add a second tool for confirmation. If your virus is anything else, the NCBI tool with custom references is a reasonable quick screen, but a tree is usually the real answer. If recombination is likely, no single rapid tool is sufficient on its own. ## Option 1: The Predefined Reference Panels (HIV-1, HCV, HBV) This is the default way to use the tool and the right starting point for the three supported viruses. For HIV-1, the panel covers the group M diversity that matters most in practice, including pure subtypes and the common circulating recombinant forms (CRFs, which are recombinant lineages with recognized names and epidemiological spread). You submit a pol or env fragment, receive a subtype call, and see immediately whether the color pattern is uniform or mosaic [1]. For HCV, the panel supports genotype and subtype calls from the sequence regions labs actually sequence. Independent evaluation work gives you a sense of how sequence-based HCV genotyping performs when anchored to NCBI resources. In one comparison of the Abbott RealTime HCV GT II assay against core and NS5B sequencing, the sequence-based calls were determined using a laboratory-developed analysis tool and the NCBI non-redundant database, and the automated assay agreed with sequencing on 95.8 percent of genotype 1 samples (96 percent for 1a, 95 percent for 1b) but only 83 percent of genotype 6 samples [2]. For HBV, the predefined panel resolves the classic genotypes A through D and beyond, which map onto serotypes and clinical features. Work on Bangladeshi HBV isolates illustrates the kind of downstream interpretation a genotype call feeds into. In that dataset, extracted from GenBank, genotypes A, C, and D made up 18, 46, and 35 percent of sequences and corresponded to serotypes adw, adr, and ayw respectively, and the analysis leaned on HBV-specific servers such as Geno2pheno-HBV and HBV Serotyper in addition to general sequence tools [6]. The pros of the predefined panels are real. There is nothing to install, nothing to curate, the references are named correctly, and the segment map answers the recombination question visually in seconds. The cons are equally concrete. Coverage stops at three viruses. Accuracy is uneven across clades, which the next section quantifies. And because the reference panel has a fixed vintage, an old snapshot of viral diversity can distort calls on newly described recombinant forms, an effect documented for the NCBI HIV subtyping implementation when it included 2009 references [7]. ## Option 2: Custom Reference Sets The ability to supply user-defined reference sequences at the same time as the built-in panels is the feature that stretches this tool far beyond its three headline viruses [1]. The workflow is straightforward in principle. You collect reference sequences with confident genotype labels, align them if the tool requires it or paste them as instructed, upload them alongside your query, and read the segment map. Custom references make sense in three situations. First, regional or recently described variants. If a recombinant form emerged after the built-in panel was frozen, your own references can carry it. Second, veterinary and zoonotic viruses. There is no built-in panel for, say, a particular pestivirus or [avian influenza](/knowledge/bacteria/avian-bacteria/avian-influenza-new-zealand-kerala-outbreaks-jessamine-county-updates) lineage, but nothing stops you from building a reference panel from GenBank records and running the same segmentation logic. Third, teaching and method development. The tool is a transparent way to demonstrate why reference choice drives genotype calls. The honest caveat is that you have inherited the curation job. Every reference needs a correct, verifiable label, sequences need to cover the same region as your query, and a panel of two representatives per genotype will give choppy, unreliable segment maps. A panel built from ten to twenty well-annotated full-length genomes per genotype behaves far better. If you cannot vouch for your reference labels, the segment map is decoration, not evidence. ## Option 3: The Rival Rapid Subtyping Tools and How They Compare ![Comparison of NCBI, Stanford and REGA accuracy on CRF02_AG, pure non-B clades and other recombinants.](/article-images/illustrations/ncbi-viral-genotyping-tool-tool-accuracy.webp) The best data on how the NCBI tool stacks up against its competitors comes from head-to-head evaluations on HIV-1, where automated subtyping is routine and the gold standard is full phylogenetic analysis. A Spanish study evaluated seven automated tools (Stanford, Geno2pheno, Rega, NCBI, EuResist, STAR, and TherapyEdge) against phylogeny on 670 pol sequences from treatment-naive patients, 587 subtype B and 83 non-B. Most tools correctly classified subtype B, but up to 15 percent of non-B sequences were wrongly identified as B depending on the tool. For subtype B and CRF02_AG, the Stanford and NCBI tools showed the highest sensitivities while Geno2pheno and Rega showed the lowest. EuResist and Geno2pheno correctly classified all 13 non-B pure subtypes at pol. Every tool dropped sharply on recombinants other than CRF02_AG, and only NCBI05, Rega, and STAR identified unique recombinant forms (URFs) at all, with very low sensitivity. Notably, the NCBI implementation classified the highest number of subtype B sequences as non-B and overestimated recombination, especially when its reference set included the 2009 references [7]. An independent Madrid-based evaluation on non-B pol sequences (277 protease and 171 reverse transcriptase sequences) tells a consistent story with sharper numbers. Two-thirds of sequences previously assigned to non-B clades were reclassified as CRFs when phylogeny was rerun. For CRF02_AG, the Stanford and NCBI tools were correct on 92 to 97 percent and 96 to 99 percent of protease-reverse transcriptase sequences respectively, while REGA managed only 18 to 32 percent. On pure non-B clades, the Stanford, NCBI, and REGA tools failed on 24 to 33 percent, 35 percent, and 57 to 64 percent of sequences respectively. For CRFs other than CRF02_AG, discrepancies hit 98 to 100 percent for Stanford, 18 to 43 percent for NCBI, and 80 to 87 percent for REGA [5]. Read those two evaluations together and a clear profile of the NCBI tool emerges. It is excellent on CRF02_AG, competitive on pure clades, one of the few tools that even attempts URFs, but prone to overcalling recombination, particularly with a 2009-era reference set. Its tendency to mislabel some B sequences as non-B is the mirror-image failure mode of the Stanford tool, which leans the other way. For HCV specifically, the rival worth knowing is COMET, an online genotyping tool that was used alongside the NCBI non-redundant database to assign genotypes in a large evaluation of the Abbott GT II assay against 5'UTR, core, and NS5B sequencing. In that study, core and NS5B sequencing resolved specimens that 5'UTR sequencing mislabeled, including two genotype 6 subtypes (6l and 6u) that both the Abbott assay and 5'UTR sequencing called genotype 1 with no subtype. Overall genotype-call concordance between sequencing and the assay was 97.6 percent, but the rare-subtype panel showed exactly where automated assays and short regions break down [3]. For HBV, the practical rivals are the HBV-specific servers noted above, Geno2pheno-HBV, HBV Serotyper, and the HIV-Grade HBV tool, which add serotype prediction and drug-resistance interpretation on top of genotype calls [6]. ## Option 4: Phylogenetic Analysis, the Gold Standard Both major evaluation papers treat phylogenetic analysis as the reference against which rapid tools are judged [5], [7]. A tree built from your query plus a well-chosen reference alignment shows you not just the closest genotype but the branching structure, and it is the only method among these options that makes recombination visible as topology rather than as a color switch. The hepatitis E virus (HEV) literature shows what the phylogenetic route looks like for a virus with no NCBI genotyping panel. Because HEV genotypes are highly heterogeneous, most analyses had been restricted to subgenomic regions amplified by PCR, an approach that can be unreliable when primer homology is poor. One team designed RNA probes against all published full-length HEV genomes in NCBI, used target enrichment before [Illumina sequencing](/blog/guides/illumina-sequencing), and recovered whole-genome consensus sequences from samples spanning viral loads from 1.25 x 10^4 to 1.17 x 10^7 IU/mL. Phylogenetic analysis of those genomes recapitulated and extended the partial-genome genotypes obtained by Sanger sequencing, confirming genotype 1 in ten samples and genotype 3 in thirty [8]. The lesson generalizes. When a virus is not among the three with predefined panels, whole-genome sequencing plus a tree is the standard route, and the NCBI genotyping tool is at most a triage step with custom references. The cost of the phylogenetic route is time and skill. You need a [multiple sequence alignment](/blog/guides/multiple-sequence-alignment), a model of evolution, a tree-building method, and the judgment to read placement honestly. For a single sequence that needs a quick label, that overhead is hard to justify. For anything you publish, it is not optional. ## Newer Options Worth Knowing About Two developments sit outside the classic tool landscape but are relevant to how you allocate effort. Machine learning approaches to HCV genotyping have matured. A 2024 study used Chaos Game Representation to map HCV nucleotide sequences into two dimensions, then applied self-supervised feature extraction with a convolutional autoencoder, reporting performance across ten genotypes and subtypes (1a, 1b, 2a, 2b, 2c, 3a, 3b, 4, 5, and 6) that outperformed various machine learning and deep learning baselines on both partial and complete genomes [9]. These are research methods, not clinical or routine tools, but they signal where sequence-based genotyping is heading, and they matter if you are building a pipeline for large HCV datasets. At the other extreme, metagenomic classifiers such as Kraken2-based pipelines solve a different problem entirely. One group built a classification tree from 715,672 metagenome-assembled viral sequences using k-mer profiles, producing a Kraken2-compatible viral database for use with ParaKraken in metagenomic and metatranscriptomic samples [10]. Use that class of tool when you do not know what virus you have. Use the NCBI genotyping tool when you know the virus and need to know the genotype. Confusing the two is a common and expensive category error. Arboviruses illustrate the boundary as well. Dengue serotyping in a 2019 Dhaka outbreak study was done by RT-PCR on 81 samples with sequencing and phylogenetic characterization of the partial C-prM gene of DENV-3 isolates [11]. Dengue has no NCBI genotyping panel, so the practical route is serotyping by PCR plus phylogenetic analysis of a standard region, with the NCBI tool's custom-reference mode as an optional screen. ## How to Choose: Five Criteria That Decide It ![Decision path asking whether the virus is HIV-1, HCV or HBV, whether recombination is likely, and what the result will be used for.](/article-images/illustrations/ncbi-viral-genotyping-tool-which-route.webp) **1. Which virus.** If the answer is HIV-1, HCV, or HBV, the predefined panels apply and the NCBI tool is a legitimate first call [1]. For anything else, you are immediately in custom-reference or phylogenetic territory. **2. Which genomic region.** Region choice changes answers, sometimes materially. HIV evaluations relied on pol sequences, and tool accuracy varies by region [5], [7]. HIV envelope analyses of the C2V3 region, converted to protein with the NCBI Translate tool, answered a different question (coreceptor tropism and glycosylation) and associated subtype with tropism at Geno2Pheno false positive rate cutoffs of 5, 10, and 20 percent for CXCR4 usage [4]. For HCV, 5'UTR sequencing mislabeled two genotype 6 subtypes that core and NS5B sequencing resolved correctly [3]. Sequence the most discriminating region your workflow allows, and never assume a subtype call from one region transfers to the whole genome. **3. How likely is recombination.** If the sample comes from a region or risk group where recombinant forms circulate, the NCBI tool's segment map is genuinely useful, because revealing mosaic structure is its design purpose [1]. But calibrate your trust. It overestimates recombination with dated references and misses most non-CRF02_AG recombinants in evaluations [5], [7]. Treat a mosaic map as a hypothesis for tree-based confirmation, not as a final call. **4. What hangs on the answer.** A quick sanity check on a sequencing run can live with a rapid tool. A genotype that will be reported to a clinician, used for epidemiology, or published needs concordance across at least two independent methods plus a tree. Individual patient decisions always belong with the treating clinician interpreting the result in context. **5. How many sequences.** The web form handles a handful of sequences at a time. For dozens to thousands, you want scripted BLAST against a curated local reference set, or one of the programmatic classifiers, with the web tool reserved for spot checks and figures. ## Worked Scenarios **Scenario A: An HIV-1 pol sequence from a West African patient, CRF02_AG suspected.** Submit the protease-reverse transcriptase fragment to the NCBI tool with the HIV-1 panel. Published evaluation data put NCBI's accuracy on CRF02_AG at 96 to 99 percent for protease-reverse transcriptase sequences, the best of the tools tested [5]. A clean, uniform CRF02_AG color block with no mosaic pattern is very likely correct. If the map shows internal color switches, reclassify through phylogeny, because recombinant structure beyond the common CRFs is where rapid tools fail [5], [7]. **Scenario B: An HCV sample with an ambiguous commercial assay result.** The evidence-based path is core and NS5B sequencing, with the genotype call made against the NCBI non-redundant database or a tool like COMET [2], [3]. Do not fall back on 5'UTR alone, because that region mislabeled genotype 6 subtypes 6l and 6u as genotype 1 in evaluation data [3]. The NCBI genotyping tool's HCV panel then serves as a fast visual cross-check on the sequencing call. **Scenario C: Hepatitis E virus genomes from serum samples.** There is no built-in panel, so build the study around whole-genome sequencing with a phylogenetic readout, as done in the HEV target-enrichment protocol that recovered full genomes across a viral-load range of 1.25 x 10^4 to 1.17 x 10^7 IU/mL and confirmed genotypes 1 and 3 by tree placement [8]. If you want a rapid screen first, run the NCBI tool with a hand-built HEV reference panel and treat the output as preliminary. **Scenario D: An HBV sequence where serotype matters.** Call the genotype with the NCBI HBV panel, then run an HBV-specific server such as Geno2pheno-HBV or HBV Serotyper for serotype and mutation interpretation. The Bangladeshi dataset shows why the pairing matters, with genotypes A, C, and D corresponding to serotypes adw, adr, and ayw and with escape-mutation patterns differing by genotype [6]. ## Step-by-Step: Running the Tool 1. **Prepare the sequence.** Start with a single, cleanly edited consensus sequence in FASTA format, oriented in the correct direction, with ambiguous bases minimized. Strip vector contamination and primer remnant. 2. **Open the tool.** Navigate to the NCBI genotyping form page (the address published with the tool is http://www.ncbi.nih.gov/projects/genotyping/formpage.cgi) or search the NCBI site for the current location [1]. 3. **Choose the reference set.** Pick the built-in HIV-1, HCV, or HBV panel, or supply your own user-defined references alongside it [1]. 4. **Set the segmentation.** The server breaks your query into segments for the comparison. Larger segments give stabler calls on uniform sequences, smaller segments resolve short mosaic tracts. Match the setting to how much recombination you expect. 5. **Submit and read the map.** The output is a color-coded graphical display along the length of your query, with each color representing a genotype, so a switch in color marks a putative recombinant breakpoint [1]. 6. **Verify before you commit.** Run BLAST against the non-redundant database as an independent check, the same way HCV reference-method studies have done [2], [3], and build a small phylogenetic tree with your sequence plus the references for anything consequential. ## Common Mistakes and Limitations **Trusting a single rapid tool on recombinants.** Every automated tool's accuracy collapsed on recombinant forms other than CRF02_AG in head-to-head testing, with discrepancies reaching 98 to 100 percent for one tool and 18 to 43 percent for the NCBI tool [5]. A second opinion is not optional for suspected mosaics. **Ignoring reference-set vintage.** The NCBI implementation overestimated recombinants when its references included the 2009 set, and it mislabeled subtype B sequences as non-B more often than any comparator [7]. Know what snapshot of diversity your panel represents, and prefer the most current references you can get. **Sequencing the wrong region.** Tool sensitivity differs by genomic region, and short or conserved regions carry real misclassification risk. The HCV 5'UTR versus core/NS5B discrepancy [3] and the dependence of HIV tool performance on the pol fragment analyzed [5], [7] both make the point. Match the region to the question. **Using the tool as a discovery engine.** The genotyping tool expects one known virus per query. Identification of unknown viruses in metagenomic samples belongs to k-mer and classifier-based platforms built for that job [10]. **Assuming universal virus coverage.** Predefined panels exist only for HIV-1, HCV, and HBV [1]. For HEV, dengue, and most other viruses, sequencing plus phylogenetics is the established route [11], [8]. **Feeding it mixed or low-quality input.** A mixed infection or a poor consensus will smear the segment map into an uninterpretable pattern. Clean up or reject the sequence first. **Reading the genotype as the whole clinical story.** Subtype can associate with phenotype, as seen between HIV-1 subtype and coreceptor tropism [4], but those associations come from specific analyses with defined tools and cutoffs. A genotype label is one data point, and individual patient care belongs with the treating clinician. ## Frequently Asked Questions ### Which viruses does the NCBI Viral Genotyping Tool support with predefined reference sets? Exactly three: HIV-1, HCV, and HBV [1]. For any other virus, you can supply your own user-defined reference sequences, but the accuracy of the call then depends entirely on the quality and labeling of the references you provide. ### Is the NCBI tool the best automated HIV-1 subtyping tool? It is the best or near-best on CRF02_AG and competitive on pure non-B clades, but no rapid tool is uniformly best. NCBI missed 35 percent of pure non-B clade calls in one evaluation and overcalled recombination, while Stanford led on some measures and REGA trailed badly [5], [7]. For publication-grade work, run two tools and a phylogenetic tree. ### Can I use the NCBI tool for HCV genotyping before treatment decisions? Use it to check and visualize a sequence-based call, not as the sole basis for treatment. In evaluation studies, HCV reference genotype calls came from core and NS5B sequencing interpreted against the NCBI non-redundant database or COMET, with agreement rates against a commercial assay of 95.8 percent for genotype 1 and 83 percent for genotype 6 [2], [3]. Treatment decisions rest on validated clinical assays interpreted by the treating clinician. ### How do I genotype a virus that is not HIV-1, HCV, or HBV? Build a reference panel from well-annotated GenBank sequences and run the tool in custom-reference mode as a screen [1], then confirm with phylogenetic analysis. The HEV whole-genome workflow, using target enrichment of all published full-length genomes followed by tree-based genotype assignment, is a worked model for this route [8]. ### Do I still need phylogenetic analysis if the tool gives me a clean genotype? If the result will be published, reported clinically, or used for surveillance, yes. Phylogeny is the gold standard both major tool evaluations were measured against, and it is the only method that reveals recombinant structure as tree topology rather than a color pattern [5], [7]. ### How long should my input sequence be, and does the region matter? There is no universal minimum, but shorter and more conserved regions give weaker and less reliable calls, and the right region differs by virus and question. HIV tool accuracy was characterized on protease and reverse transcriptase fragments [5], while HCV core and NS5B outperformed 5'UTR for subtype-level resolution [3]. Sequence the most informative region your protocol allows and report which region you used alongside the genotype call. ## Sources 1. [A web-based genotyping resource for viral sequences.](https://pubmed.ncbi.nlm.nih.gov/15215470/) 2. [Evaluation of the Abbott RealTime HCV genotype II plus RUO (PLUS) assay with reference to core and NS5B sequencing.](https://pubmed.ncbi.nlm.nih.gov/28324789/) 3. [Evaluation of the Abbott realtime HCV genotype II RUO (GT II) assay with reference to 5'UTR, core and NS5B sequencing.](https://pubmed.ncbi.nlm.nih.gov/24656214/) 4. [Partial HIV C2V3 envelope sequence analysis reveals association of coreceptor tropism, envelope glycosylation and viral genotypic variability among Kenyan patients on HAART.](https://pubmed.ncbi.nlm.nih.gov/28196510/) 5. [Reliability of rapid subtyping tools compared to that of phylogenetic analysis for characterization of human immunodeficiency virus type 1 non-B subtypes and recombinant forms.](https://pubmed.ncbi.nlm.nih.gov/18842935/) 6. [A meta-analysis on genetic variability of RT/HBsAg overlapping region of hepatitis B virus (HBV) isolates of Bangladesh.](https://pubmed.ncbi.nlm.nih.gov/31709005/) 7. [Sensitivity of seven HIV subtyping tools differs among subtypes/recombinants in the Spanish cohort of naïve HIV-infected patients (CoRIS).](https://pubmed.ncbi.nlm.nih.gov/21070813/) 8. [Hepatitis E virus: Whole genome sequencing as a new tool for understanding HEV epidemiology and phenotypes.](https://pubmed.ncbi.nlm.nih.gov/33933822/) 9. [On leveraging self-supervised learning for accurate HCV genotyping.](https://pubmed.ncbi.nlm.nih.gov/38965254/) 10. [A k-mer based approach for classifying viruses without taxonomy identifies viral associations in human autism and plant microbiomes.](https://pubmed.ncbi.nlm.nih.gov/34849195/) 11. [Clinical and laboratory comparison of severe (Group B and C) Dengue cases with molecular characterization from 2019 epidemics in Dhaka, Bangladesh.](https://pubmed.ncbi.nlm.nih.gov/39585911/)