The Nextstrain team has been focused on extending the usability of Auspice and\nauspice.us. Here are a few features that we wanted to highlight:
\nWe have released a new version of auspice.us that allows the site to be\nused offline without an internet connection. After opening the site once with\nan internet connection, you should be able to do the same while offline using the same browser.
\nNote this does not support the map panel since it requires internet connection to\ndownload map tiles.
\nWhen choosing to download a JSON from within Auspice we now create the JSON\ndynamically rather than re-fetching it from the underlying source. This allows\nauspice.us to use this functionality, including when\nstarting from newick trees, as well as containing any dragged-on metadata.\nWe attempt to reflect the current UI state in the downloaded dataset, so that\nthings like the current color-by, tree layout etc become the new defaults in\nthe downloaded JSON; in the future we will extend this to (e.g.) use the\ncurrent zoom state to download the subtree.
\nWe have released a new version of auspice.us that allows users to edit a subset\nof dataset metadata and colorings within the app. The changes are saved locally\nin the brower and users must download the Auspice JSON to save the changes.\nPlease see our docs on Editing Datasets\nfor more details.
\nThe drag-and-drop metadata feature has been made more powerful by merging data\ninto existing colorings and supporting node data JSON files. In addition, the\nmerged metadata is now included in the downloaded Auspice JSON so that users\ncan keep a local copy of the merged dataset. For more details, please see the\ndrag-and-drop extra metadata\ndocs page.
\nMost Nextstrain datasets focus on nucleotide analyses, but we've had requests\nfrom users for support of entirely amino acid analyses.\nAs of Augur 33.1.0 &\nAuspice 2.69.0\nwe can now run such analyses. Please see our\nAmino Acid Workflows guide\nfor more details.
\nNextstrain strives to make all of our data open and available and thus has made\nthe derived data behind visualizations available for download through Auspice.\nWe recognize that other data sources may not permit open data sharing, so\nwe've made it possible for users to control which download options are available\nfor their Auspice dataset. Please see our\nsharing controls\nfor more details.
\n", "url": "https://nextstrain.org/blog/2026-07-09-auspice-updates", "title": "Auspice and auspice.us updates", "date_modified": "2026-07-09T00:00:00.000Z", "author": { "name": "The Nextstrain team" } }, { "id": "https://nextstrain.org/blog/2026-06-24-new-location-for-discussion-forum", "content_html": "Our discussion forum has moved from discussion.nextstrain.org to nextstrain.discourse.group.
\nYou can continue to use your existing account at the new location. Old links will be automatically redirected.
\nPlease contact us if you experience any issues.
\n", "url": "https://nextstrain.org/blog/2026-06-24-new-location-for-discussion-forum", "title": "New location for discussion forum", "date_modified": "2026-06-24T00:00:00.000Z", "author": { "name": "The Nextstrain team" } }, { "id": "https://nextstrain.org/blog/2026-05-19-annual-update-may-2026", "content_html": "Nextstrain development supports real-time genomic surveillance for two central purposes:
\nOver the past year we advanced these goals through continued development of pathogen specific resources, targeted outbreak response, and advances to the bioinformatic tooling and infrastructure needed to support real-time genomic surveillance.
\nThis year we continued to track the evolution and spread of important seasonal respiratory viruses. Our SARS-CoV-2 and seasonal influenza builds automatically tracked the emergence and spread of novel, rapidly expanding lineages such as BA.3.2 in SARS-CoV-2 and subclade K in influenza H3N2. The recent H5N1 panzootic is a pertinent reminder that seasonal antigenic drift makes up a small portion of the global evolutionary dynamics of influenza. We restructured the underlying data for our influenza builds to better support segment specific analyses that span the full host-range of the virus.
\nWe responded to two emerging outbreaks this past year. In response to the 2025 Ebola outbreak in Kasai (DRC) we developed a unified analysis that places this outbreak in context with all other Ebola outbreaks. We also updated our measles builds (whole genome and N450) to better capture the recent outbreak spreading in the US and have provided a USA-specific analysis as well.
\nWe continued to add baseline analyses across endemic and emerging viral pathogens. For each pathogen we provide a GitHub repository that contains workflows for automated data ingest from NCBI GenBank, a pipeline to generate Nextclade datasets, and automated phylogenetic analysis. This year we completed automated builds for
\nIn particular we would like to highlight our work with M. tuberculosis. TB marks Nextstrain's first bacterial core pathogen. Not only does this work expand our impact beyond viruses, but the process of developing an automated TB build facilitates using FASTQ as entrypoint for Nextstrain workflows, consistent with community standards for sharing bacterial sequencing data.
\nWe've posted a preprint to bioRxiv describing our computational pipelines and infrastructure for data curation and automated rebuilds for 21 viruses and M. tuberculosis. It's available as:
\nThis year we introduced several advances to our tool chain that make reproducible phylogenetic analyses more user friendly. Over the summer we introduced nextstrain run, a 'git-free' way to install and run Nextstrain pathogen workflows. The run command separates input files and results from the often messy pipeline code needed for a reproducible pathogen build. Users are able to customize pipeline arguments through a config file without knowledge of Snakemake. Nextstrain run marks a key step in increasing the accessibility of genomic surveillance and we look forward to including this feature in future workflows.
We standardized and simplified the way users can combine their own data with contextual data in a pathogen build. The pattern allows for specification of multiple inputs to a workflow with a common setup being a base input of curated data from data.nextstrain.org and \"additional input\" of local user data. These data are intelligently combined so that user data can act as an overlay to the base of curated data.
\nSelecting appropriate sequence data for an analysis often requires subsampling, and we introduced augur subsample and augur proximity to help with this process. Our new augur proximity command enables assembling a contextual dataset by selecting sequences genetically close to a focal dataset. Augur subsample allows for more involved strategies than were previously possible with augur filter. The subsamples can be defined in a config file. With the addition of proximity and subsample it is now easier to design complex or hierarchical subsampling schemes through YAML configuration without the need to write bespoke Snakemake rules.
We also made several improvements to our phylogenetic visualization software Auspice. We shipped the \"streamtrees\" feature that was alluded to last year. These collapsed clades condense large phylogenetic trees into more manageable components while still displaying the frequency of traits in each clade. We introduced a 'focus on selected' feature, which clearly distinguishes regions of the tree that are ancestral to selected tips. This view approximates a build with just the selected samples allowing for rapid exploration of large datasets. More recently, we extended the feature to also horizontally zoom into parts of the tree that cover a specific time span. Finally, we implemented a new dataset modal selector to streamline navigation between datasets available for a particular pathogen. These visualization developments are expanded on in a recent blogpost.
\nWe continued to improve Nextclade in line with our roadmap from last year. In particular,
\nAn integrated support for segmented viruses has not yet been tackled. We also maintained and expanded the available datasets. Mpox now supports the new WHO recommended nomenclature, influenza datasets now exist for all segments and all human lineages. In addition to the datasets we maintain, we have also seen encouraging uptick in community contributions. Emma Hodcroft's group started a consortium collection for Enteroviruses, Louise Moncla's group maintains avian influenza H5 datasets, ITPS and V-Gen Lab all contributed arbovirus datasets.
\nOver the past year we also worked to increase the transparency and reproducibility of Nextstrain hosted analyses. We have begun to rely more heavily on open source data and rolled out support for Pathoplexus as a data source where applicable. Currently, we use Pathoplexus data in mpox, RSV, Ebola, measles, HMPV and WNV analyses. Where feasible, we also surfaced the sequence and metadata files behind each build at nextstrain.org/pathogens/files. These resources provide users with direct access to the data used in each build and can serve as a starting point for their own analyses. This philosophy is informed in part by GISAID's unilateral decision to terminate SARS-CoV-2 datafeed for Nextstrain and other groups.
\nWe scaled back on our previous plans to develop a fully fledged \"frequencies\" app. Instead in the past year, we continued to develop our approach to clade/lineage frequency forecasting of a pre-defined set of clades or lineages. This work can be seen for seasonal influenza virus at nextstrain.github.io/forecasts-flu/.
\nDuring the 2026-2027 period we plan to work towards increasing the accessibility of genomic surveillance to experts and non-experts alike.
\nFunding for Seattle-based operations for 2026-2027 will come primarily through:
\nFunding for Basel-based operations for 2026-2027 will come primarily through:
\nNextstrain started as a collaborative project coded up by Richard Neher and Trevor Bedford in early 2015, targeted exclusively to seasonal influenza tracking, then called nextflu. Since then it's grown to encompass a number of pathogens and has had a number of contributors from both Richard's lab and Trevor's lab. We've also seen a robust community of users who've contributed analyses via Groups and via GitHub. We believe our central approach has advanced and continues to advance actionable genomic surveillance.
\nAfter more than a decade leading the Seattle side of Nextstrain, Trevor is stepping back from day-to-day operations to pursue new scientific directions focused on deep learning and biological sequence modeling. JT McCrone will be taking over from Trevor as PI for the Seattle-based portion of Nextstrain run out of the Fred Hutch. JT will serve as PI on both the Gates Foundation and CDC awards going forward. Long time contributors James Hadfield, Jover Lee and Victor Lin are remaining under JT's supervision in the Seattle office as direct Nextstrain contributors. John Huddleston (along with Richard) is still leading work on seasonal influenza. Ivan Aksamentov and Cornelius Roemer are remaining under Richard's supervision in the Basel office. Emma Hodcroft remains an independent contributor in her new role as PI at Swiss TPH. JT and Richard will co-lead Nextstrain going forwards, with Trevor continuing to attend calls and serving in an advisory capacity.
\n", "url": "https://nextstrain.org/blog/2026-05-19-annual-update-may-2026", "title": "Nextstrain Annual Update May 2026", "date_modified": "2026-05-19T00:00:00.000Z", "author": { "name": "JT McCrone, Richard Neher, Trevor Bedford" } }, { "id": "https://nextstrain.org/blog/2026-03-12-auspice-updates", "content_html": "With more and more genomic data available the size of phylogenetic trees being generated are growing ever larger.\nThis poses immediate problems for traditional phylogenetic tree representations as we have more tips to display than pixels to display them in, not to mention the computational costs involved.\nImportant work in this area has been done by a number of tools, especially Taxonium.\nRecent work in Auspice has been focused on displaying larger datasets whilst still conveying the important structure of the tree.\nThis has manifested in two big changes, which we describe in this blog post.
\nAs trees become bigger in both dimensions (e.g. temporally and number of tips), we want to see the parts of the tree we're interested in both in the context of the overall tree and drill down into them. In a 400-tip tree this is generally not an issue, but in a 4,000-tip tree it can become difficult to highlight the parts you wish to.
\nThere's three ways we can select nodes of interest:
\n\n\n
As you perform these actions the byline (above the tree) will indicate how many tips you have selected: \"Showing XX of YY genomes ...\".
\nAnd a few different ways we can zoom into those selected samples:
\n\n\n
P.S. Clicking \"Zoom to root\" (top right of the tree panel) returns us back to the entire tree view.
\nLet's use our 12y influenza H3N2 tree as an example (Figure 1, top left).\nImagine we're interested in a few clades which looked like they were growing but eventually died out (Figure 1, top right).\nPreviously it's been really hard to focus on these strains, as they represent ~15% of the total tips and thus occupy a similarly small fraction of the available space; and since the clade which includes them all is essentially the entire tree, \"zoom to selected\" doesn't help much here!\n(Selecting an early time slice in the tree was another common situation with the same issues.)
\nToggling on \"focus on selected\" in the sidebar allows the selected strains to fill the viewport (Figure 1, bottom).\nThis new mode rescales the tree in both dimensions: in the horizontal direction we change the viewport to only show the temporal / divergence range encompassed by the selected nodes; in the vertical dimension we recompute the position of selected tips so that they occupy ~80% of the available space, and compress the unselected tips into the remaining space (we also don't show most of the tips above & below the selected ones).
\n\n
![\"Figure1\"](\"/blog/img/auspice-updates-2026-focus-on-selected.png\")
\n \nFigure 1. (top left) our 12-year influenza H3N2/HA tree; (top right) selecting a few clades of interest often results in them occupying only a small portion of the viewport, with no easy way to zoom into just those clades; (bottom) toggling \"focus on selected\" brings these strains front and center, maximizing their appearance in the viewport.\nColor indicates influenza clade and black crosses indicate vaccine strains.
\nFocus on selected is a mode, so while it's toggled you can make changes to the selected strains (filters, date sliders, zooming into clades) and the layout will update so that the (newly) selected strains continue to take up most of the available space. Sometimes it's helpful to toggle it off to see the selected strains in the context of the overall tree and then back on to focus on just those strains.
\nFocus on selected has been around in one form or another since version 2.59.0 (October 2024). Originally it simply zoomed in the vertical dimension, but since version 2.68.0 (January 2026) it zooms in both the vertical and horizontal dimension.
\nRather than iterating on traditional tree display approaches, streamtrees break a phylogenetic tree up into partitions and display each partition as a self-contained streamgraph, replacing the branching patterns in the tree with a summary of the strains involved.\nThis allows a higher-level overview of the pathogen's evolution, focusing on which parts of the evolutionary tree are growing and how they are related to each other, whilst still conveying how metadata traits are distributed across these trees.
\nStreamtrees is a visualisation mode which can be toggled on in the sidebar similarly to the \"focus on selected\" mode discussed above.\nYou can go between streamtree view and normal tree layout with this toggle; this is especially useful when you've zoomed into a single streamtree (we may make this the automatic behaviour in the future).
\n![\"How](\"/blog/img/auspice-updates-2026-streams-explained.png\")
\n\nFigure 2: How streamtrees work. The left hand side shows a traditional rendering of the West African Ebola outbreak (2013-16), as part of our all-outbreaks Ebola summary dataset; colours represent Country and horizontal axis is genetic divergence. Branch labels divide these tips into the main outbreak clade (Ebov-2013) and a relapse clade (Ebov-2013/r2021, red box). \nStreamtrees (right hand side) use these branch labels to partition the data and draw each partition as a set of stacked ribbons, where each ribbon (i.e. each colour) represents the current selected metadata trait. Connecting branches link these streamtrees together, here indicating that the relapse clade is a child of the main outbreak clade.\nHovering on individual colours (ribbons) in each graph shows more information, and clicking on branches zooms in as normal.
\nBehind the scenes, streams are a kernel density estimate (KDE) where each tip is approximated with a Gaussian. This allows any metadata (coloring) available in the tree to be represented in streams. We go into more technical detail about the actual implementation in the Auspice docs.
\nHow tips are partitioned (where each partition becomes a streamgraph) is up to the analysis pipeline, and datasets without branch labels won't be able to make use of streamtree visualisation.\nUsing genetic clades or lineages is a natural approach for many viruses, e.g. seasonal-influenza and SARS-CoV-2; using augur clades will create branch labels for you by default and you can customise the branch-label key via --label-name.\nGeographic jumps are another common way to split up the tree, and augur traits now has a --branch-labels option to do this.\nIn other cases, custom code may be employed such as for the all-outbreaks Ebola tree in Figure 3:
\n
![\"all](\"/blog/img/auspice-updates-2026-streams-ebov.png\")
\n\nFigure 3: Streamtree of all (sequenced) Ebola outbreaks The main figure conveys the different outbreaks and the relationship between them (horizontal axis: genomic divergence, colour: sampling year). You can zoom into different streams, and toggle between streamtrees & regular layout, as done in the bottom-right inset where we have zoomed into the recent 2025 outbreak in the DRC and want to see the detailed relationships between strains.\nNote: The dataset has two different labellings for outbreaks, here we're viewing geographic names but there is also a non-geographic labelling.
\nPerformance matters
\nStreamtrees come into their own when looking at very large trees as they don't suffer from the same scaling problems as regular trees, both in terms of pixels and performance (Figure 4).\nTaxonium's approach - a sparsification algorithm that renders a dynamically-chosen subset of nodes depending on your zoom level - is another direction to solving these problems which works stunningly well for massive trees such as millions of SARS-CoV-2 genomes. \n(P.S. Nextstrain datasets are viewable in Taxonium - click the \"view in other platforms\" button at the bottom of the page in Auspice to get there.)\nWe chose to explore streamtrees for two reasons, firstly we wanted to go beyond traditional tree rendering approaches and try to give a better big-picture overview of the evolution we are capturing, and secondly as partitioning the data opens the door to partitioning the analyses which is well suited to our bioinformatics tools and pipelines.
\n\n
![\"big](\"/blog/img/auspice-updates-2026-streams-big-trees.png\")
\n\nFigure 4: Big trees work well as streamtrees Streamtrees allow us to render large trees whilst maintaining performance.\nWe think they provide a better way to convey relationships between important attributes in the data (e.g. clades), whilst allowing you to see the fine-grained tree structure on-demand.\nTop left: 17k dengue virus strains partitioned by DENV genotype, with colour indicating geographical region, which clearly separate out the different labels which are useful for outbreak tracking; top right: the typical rendering of the same 17k tree.\nBottom left: 23k SARS-CoV-2 samples partitioned by nextstrain clade, with sampling focused on King-County (WA, USA).\nThe streamtree (bottom left) is simpler to understand the overall patterns in the data and interactivity remains performant whereas rendering every node (bottom right) has degraded performance and obfuscation problems meaning the big-picture relationships tend to be drowned out by the sheer number of nodes on display; dataset created as part of Paradis et al.
\nFuture directions\nThere's many places we'd like to take streamtrees as they open up interesting visualisation ideas as well as changes to how we may analyses pathogens, especially for those where we have large amounts of sequences available.
\nJT McCrone & Andrew Rambaut used a fishplot-like approach to representing uncertainty in a SARS-CoV-2 tree, and these may be a nice way to represent the relationship between streamtrees rather than the traditional branches we currently use. \nSimilar approaches have been used for clonal evolution of cancer cells, e.g. Sandmann et al.
\nStreamtrees should also scale well to more diverse datasets, such as all-influenza datasets, where streamtrees could be used to summarise the big picture relationships (e.g. between subtypes) which then transition to a more fine-grained analysis (e.g. within subtype) as we zoom in.\nPartitioning datasets like this naturally allows us to split one giant analysis into multiple parts, which has benefits (parallisation) but also complications (linking, sharing data between partitions).\nExploring these directions is a focus for us over the coming years.
\nStreamtrees were introduced in Auspice version 2.63.0 (June 2025).\nThey are still experimental, and feedback is welcome!\nMore technical information is available in the Auspice docs.
\nSince our Nov 6 update, we haven't had further correspondence with GISAID Secretariat and the GISAID data feed for SARS-CoV-2 hasn't been updated.\nWe believe it's likely safe to assume that the feed won't be returning and so our GISAID-based analyses at nextstrain.org.ncov/gisaid will remain on hold indefinitely with a last update on Oct 1, 2025.\nMeanwhile, we continue to update the open analyses based on GenBank / INSDC data weekly to produce phylogenetic analyses at nextstrain.org.ncov/open as well as variant fitness analyses at nextstrain.org/sars-cov-2/forecasts.
\nMeanwhile, on Jan 20, GISAID posted a statement that obliquely refers to Trevor Bedford as \"individual four\" and states that \"GISAID terminated the data feed provided to a developer of software for visualizing phylogenomic data, after a review revealed that 'regional data packages' generated by the developer and offered through GISAID, were for the most part no longer being downloaded.\"\nAs we described in detail in our Nov 6 blog post, we contend that making metadata and sequences available through the \"nextregion\" interface was a secondary byproduct enacted in May 2020 to allow reproducible research given strictures of GISAID's data resharing policy.\nWe had previously documented emails from Jan 2020 and May 2020 that show this to be the case.\nThis re-writing of history matches previous behavior by GISAID regarding their contentions surrounding the first publicly shared SARS-CoV-2 genome sequence.
\nCOG-UK, outbreak.info and Cov-Spectrum were also referenced in GISAID's Jan 20 statement.\nConsequently, we, along with members of COG-UK, outbreak.info and Cov-Spectrum, have issued a joint response to GISAID that can be found at github.com/andersen-lab/2026_gisaid_response.
\n", "url": "https://nextstrain.org/blog/2026-02-24-gisaid-joint-response", "title": "Joint response to GISAID regarding termination of SARS-CoV-2 data feeds", "date_modified": "2026-02-24T00:00:00.000Z", "author": { "name": "Trevor Bedford, Richard Neher and the Nextstrain team" } }, { "id": "https://nextstrain.org/blog/2026-01-14-norovirus", "content_html": "Historically called \"winter vomiting disease\" (Adler and Zickl, 1969), norovirus is the bane of any parent, long term care facility, or otherwise contained community. The name of the virus is a mangled reference to Norwalk, Ohio, where a particular 1968 outbreak surged through a population of school children within a 24 hour period. Despite its explosive symptomology, the etiological agent was not identified as a virus until 1972, by immune electron microscopy (Kapikan et al, 1972) and not genetically sequenced until 1989 (Jiang et al, 1993). The subsequent classification of the virus was fraught as it became clear that the virus undergoes recombination, frequently between the ORF1 and ORF2 region (Bull et al, 2005), resulting in different evolutionary histories between the polymerase (RdRp in ORF1) and surface capsid (VP1 in ORF2). Despite multiple attempts and discomfiture among those afflicted, there is no approved norovirus vaccine.
\nNextstrain provides regularly updated phylogenetic monitoring of norovirus along several different facets. Since this is a highly recombining virus, we provide individual gene trees as well as the full genome tree of all norovirus samples, building off the prior effort of Allison Li and Katie Kistler and John Huddleston. They have also faceted the genome trees along important VP1 types GII.2, GII.3, GII.4, GII.6 and GII.17. Therefore, as of the time of this writing, we provide 14 regularly updated views of norovirus evolution.
\n| group | \ngenome | \np48 | \nNTPase | \np22 | \nVPg | \n3CLpro | \nRdRp | \nVP1 | \nVP2 | \n
|---|---|---|---|---|---|---|---|---|---|
| all | \ngenome | \np48 | \nNTPase | \np22 | \nVPg | \n3CLpro | \nRdRp | \nVP1 | \nVP2 | \n
| GII.2 | \ngenome | \n\n | \n | \n | \n | \n | \n | \n | \n |
| GII.3 | \ngenome | \n\n | \n | \n | \n | \n | \n | \n | \n |
| GII.4 | \ngenome | \n\n | \n | \n | \n | \n | \n | \n | \n |
| GII.6 | \ngenome | \n\n | \n | \n | \n | \n | \n | \n | \n |
| GII.17 | \ngenome | \n\n | \n | \n | \n | \n | \n | \n | \n |
The combination of highly diverged and recombined sequences proved a challenge in rooting the phylogenetic trees, and we advise that any results should be interpreted with caution. Even so, the Nextstrain trees are provided as a broad summary of the genetic diversity and relatedness, and further biological interpretation may require targeted sampling, tuning of parameters, a different alignment reference, or focusing on particular gene combinations. The trees are being annotated by both VP1 and RdRp types. From the map, norovirus types do not appear to be geographically segregated. From the frequency panel, we see indications that there are dynamics of leading types and it is not a virus that has reached genetic equilibrium of the proportion of those types.
\n![\"Figure](\"/blog/img/norovirus-all-genome-view.png\")
\nFigure 1. Norovirus is globally distributed and highly divergent. Phylogenetic trees are annotated by both VP1 and RdRp types, host, country, date, genome and gene coverage percentages. The full genome tree is shown here. From the map, norovirus types do not appear to be geographically segregated. From the frequency panel, we see indications that there are dynamics of leading types and it is not a virus that has reached genetic equilibrium of the proportion of those types.
Norovirus samples have a dual-typing system based on a polymerase region (RdRp) and capsid region (VP1) of the genome, between which is a known recombination site. The resolution of norovirus typing has undergone multiple changes (Zheng et al., 2006; Eden et al., 2013; Chhabra et al., 2019; Tatusov et al., 2020), but generally are split into a \"genogroup\", \"genotype\", and \"variant\" classification for VP1, and \"P-group\", \"P-type\", and \"variant\" for RdRp. For the sake of naming Nextstrain trees, we will name these VP1 group, type, or variants and RdRp group, type or variants respectively.
\n![\"Figure](\"/blog/img/norovirus-group-type-variant.png\")
\nFigure 2. Typing of norovirus samples is based on the VP1 and RdRp region and are further split out into group, type, and variant resolution.
The public Nextstrain metadata includes group, type, and variant levels of resolution that were roughly classified by a preliminary Nextclade datasets based on either VP1 or RdRp gene and should be used with caution. The Norovirus Nextclade datasets have not been published yet, pending further fine tuning to reduce false positives of GII.4. These datasets have been built from scaffold strains listed at the Human Calicivirus Typing Tool as of September 16, 2025.
\n![\"Figure](\"/blog/img/norovirus-nextclade-results.png\")
\nFigure 3. Preliminary norovirus classification into Group, Type, and Variant columns.
We curate sequence data and metadata from NCBI as the starting point\nfor our analyses. We provide snapshots of the exact curated sequences and metadata for norovirus workflows at:
\n\nWe welcome comments or suggestions from norovirus researchers to improve these Nextstrain and Nextclade datasets. Special thanks for feedback from Chao-Yang Pan and Erik Wolfsohn for answering questions and providing some biological context.
\n", "url": "https://nextstrain.org/blog/2026-01-14-norovirus", "title": "New Resources for Norovirus", "date_modified": "2026-01-14T00:00:00.000Z", "author": { "name": "Jennifer Chang, Jover Lee, Kim Andrews, James Hadfield, Allison Li, Katie Kistler, John Huddleston, Trevor Bedford" } }, { "id": "https://nextstrain.org/blog/2025-11-17-tuberculosis-resources", "content_html": "Nextstrain has now released continually updated genomic surveillance resources for Mycobacterium tuberculosis, the bacterium that causes tuberculosis (TB). Results of these resources are available at nextstrain.org/tb/global (Fig. 1).
\nTB is a major global health issue, causing more deaths around the world than any other infectious disease (WHO Global tuberculosis report 2024). About a quarter of the world's population is estimated to have been infected with M. tuberculosis, although in most cases the infection is latent, with only 5-10% of infected individuals expected to fall ill. TB is curable through antibiotic treatment, but the emergence of drug resistant strains of M. tuberculosis has complicated our ability to control this disease on a global scale.
\nOur new Nextstrain M. tuberculosis genomic surveillance resources aim to contribute to global monitoring of this pathogen by providing continually updated phylogenetic analyses using publicly available sequence data for strains from across the world. Using community-provided tools, we also predict whether strains are drug resistant and classify strains by phylogenetic lineage (Fig. 1). The analysis is updated every week using a random subset of approximately 1000 M. tuberculosis samples from across the world over time from the NCBI SRA. The phylogeny includes a random subsample from all lineages in the M. tuberculosis complex that have available sequence data, including both human-adapted and animal-adapted lineages. Mycobacterium canettii is excluded from the phylogeny due to its high genetic divergence from other lineages.
\nA. Drug resistance types\n![\"Figure](\"/blog/img/tb-fig1a.png\")
\nB. Lineage assignments\n![\"Figure](\"/blog/img/tb-fig1b.png\")
\nFigure 1. Phylogenetic tree of Mycobacterium tuberculosis samples from across the world alongside a map showing distribution of the samples, available at nextstrain.org/tb/global. Samples can be colored by A) predicted drug resistance and B) Lineage assignments.
These M. tuberculosis genomic surveillance resources represent Nextstrain's first real-time analyses for a bacterial pathogen. Although these resources share the same outputs as our viral pathogen resources, the underlying analyses differ in many ways to accommodate the distinct genomic characteristics of bacterial pathogens in general, and M. tuberculosis in particular.
\nOne of the main differences is that the workflow starts from raw Illumina sequence reads for each sample, whereas our viral workflows start from genome assemblies (Fig. 2). The raw sequence reads are aligned to a reference genome and are then used to identify variable sites within each genome using the program snippy. Variable sites are then summarized in a VCF file that is used as input for the phylogenetic analysis. This is in contrast to our viral workflows, for which the phylogenetic input is a full genome alignment in a FASTA file.
\n![\"Figure](\"/blog/img/tb-fig2.png\")
\nFigure 2. Comparison of analysis steps in Nextstrain real-time genomic surveillance resources for a typical viral workflow versus the Mycobacterium tuberculosis workflow, focusing on steps conducted prior to phylogenetic analysis.
Our M. tuberculosis workflow also uses the program TBProfiler (Phelan et al. 2019, Verboven et al. 2022) to predict resistance to anti-tuberculosis drugs by comparing the genome sequences of each sample against the tbdb reference database, which has a list of mutations associated with drug resistance published by the World Health Organization and other sources (Fig. 2). TBProfiler also assigns a phylogenetic lineage to each sample using a reference database of lineage-specific mutations. Drug resistance and lineage classifications are provided as options for coloring on the phylogeny.
\nThe M. tuberculosis workflow requires substantially more computational resources than most of our viral workflows due to 1) the much larger genome size of bacteria compared to viruses, and 2) the use of raw sequence read files, which are much larger than genome assembly files. One way we address these computational requirements is by using Amazon Web Service (AWS) high-performance computing resources for our weekly runs. Every time we run the analysis, TBprofiler and snippy results are generated and stored on AWS for each of the samples that was randomly selected for the analysis. The next time the analysis runs with a new random subsample, the TBprofiler and snippy results for any sample that was previously analyzed are downloaded without having to re-run TBprofiler and snippy.
\nWe gratefully acknowledge the authors, originating and submitting laboratories of the genetic sequences and metadata for sharing their work. Additionally, we welcome comments or suggestions from TB researchers on how to improve these Nextstrain datasets for their use case.
\n", "url": "https://nextstrain.org/blog/2025-11-17-tuberculosis-resources", "title": "New Resources for Tuberculosis", "date_modified": "2025-11-17T00:00:00.000Z", "author": { "name": "Kim Andrews, James Hadfield, Victor Lin, Jover Lee, Jennifer Chang" } }, { "id": "https://nextstrain.org/blog/2025-11-06-gisaid-based-ncov-analyses", "content_html": "On Oct 1, 2025, we received an email from GISAID Secretariat informing us that GISAID has immediately ended updates to the flat file of SARS-CoV-2 genomic sequences and associated metadata that they had provisioned to Trevor Bedford for updating Nextstrain analyses since Feb 2020.\nGISAID's stated rationale was that their \"resources are limited\".
\nThis file was initially provisioned Jan 30 2020, and we had kept nextstrain.org/ncov updated from it to show the genetic relationships among all SARS-CoV-2 viruses sequenced and shared to GISAID.\nAs the dataset grew, we developed a subsampling strategy that faceted the data by different continent-level regions, by different recency windows (ie 2 months back, 6 months back or the entire timeseries) and by different rootings (ie to Wuhan or to clade 21L).\nOn GISAID's request, we made specific changes to the website to do things like surface GISAID's logo at the top of the page, to keep the hCoV-19/ prefix in strain names and to disable download of associated metadata.\nWe also arranged with GISAID for them to host matching sequence and metadata files because we're not permitted to reshare GISAID data under their Database Access Agreement.\nStarting May 2020, these were available through the \"nextregion\" section of \"Genomic Epidemiology\" in their EpiCoV interface.
\nMore precisely, in ending updates to our flat file, GISAID stated that our 'nextregion' package was not being downloaded with a \"frequency that justifies the effort required by GISAID to package and prepare the information\".\nOur belief is that the primary public health utility and original purpose of the data feed from Jan 2020 was to enable continued updates to nextstrain.org/ncov, which still sees significant daily visitors.\nMaking metadata and sequences available through the \"nextregion\" interface was a secondary byproduct enacted in May 2020 to allow reproducible research given strictures of GISAID's data resharing policy.
\nIn October's email exchange, we requested continued flat file access to keep nextstrain.org/ncov updated, but GISAID has refused this request stating that \"after consulting with our staff and advisors on the feasibility of keeping your global tree up-to-date, there was a clear consensus that continuing to generate, zip and move big files back and forth is not sustainable and a waste of resources.\"\nThis has not made sense to us as GISAID can keep a single flat file up-to-date and available across analysis websites including LANL, CoV-Spectrum, Nextstrain, etc...\nThere are also simple technical revisions (for example sharding by submission year) that would easily reduce resource overhead in generating this single file for use across external analysis websites.
\nInstead GISAID has proposed that we \"provide to GISAID the parameters for Augur and GISAID will run it for you. You would then be provided the Augur output JSON file with the relevant subsample for your phylogenetic tree.\"\nHowever, our analyses pipelines are developed continuously and we don't think this proposal is a viable or desirable solution.\nWe believe that continued flat-file access is the appropriate modality and easiest way forward to keep genomic surveillance operating.\nAdditionally, we think it is important to keep a bright line between analyses run by Nextstrain and those run by others.\nThat said, we're open to finding a solution that would respect this, with a conceivable option of GISAID produced JSONs shared directly by GISAID through a new Groups page.\nWe don't have a timeline or details on continued GISAID-based analyses at the moment.
\nWe firmly support labs collecting specimens and generating sequence data and strive to credit these labs prominently.\nBut credit for data contributions does not need to be a zero sum game – appropriate surfacing of data in popular tools generates visibility and recognition and does not infringe on future publications by the data generators.\nReal-time genomic surveillance including early variant warning requires a healthy analysis ecosystem.\nTools like Nextstrain, CoV-Spectrum, UShER and outbreak.info facilitate a global community of experts to keep close tabs on the ongoing evolution of SARS-CoV-2.\nSupport for Outbreak.info ended in Jan 2025 and CoV-Spectrum hasn't been updated for >3 weeks.\nClosing these public analyses puts the world more in the dark and concretely harms surveillance as it becomes more difficult for variant spotters to contribute to global situational awareness.\nEnding flat-file access undermines vital surveillance infrastructure and conflicts with GISAID's mission to promote rapid data sharing and develop new research tools.
\nOur open analyses based on GenBank / INSDC data continue to operate as usual.\nThere is a fair amount of recent data here, but it is geographically restricted as many countries submit primarily to GISAID.\nIf you're interested to see your data in these open analyses please consider submitting sequences to INSDC via the NCBI SARS-CoV-2 submission portal.
\nWe believe transparency is essential in these matters of global genomic surveillance.\nAs such, we requested a fact check from GISAID prior to publication of this blog post.\nOur queries, GISAID's verbatim responses and our clarifications are available in this PDF document.
\n", "url": "https://nextstrain.org/blog/2025-11-06-gisaid-based-ncov-analyses", "title": "Interruption to GISAID-based SARS-CoV-2 sequence analyses", "date_modified": "2025-11-06T00:00:00.000Z", "author": { "name": "Trevor Bedford, Richard Neher and the Nextstrain team" } }, { "id": "https://nextstrain.org/blog/2025-09-29-standardized-multiple-inputs", "content_html": "The Nextstrain team is continuing our push to facilitate running pathogens\nworkflows with user data and user config as outlined in our planned directions for 2025.\nWe have decided to standardize configuration parameters for defining multiple\ninputs for phylogenetic workflows and have updated our pathogen-repo-guide with\nthe latest guidance.
\nOur phylogenetic workflows will define the default inputs for Nextstrain,\nwhich usually links to Nextstrain curated data produced from our ingest workflows.\nCustom builds are then expected to include their own inputs with additional_inputs\ndefined in the config.yaml.
inputs:\n - name: nextstrain\n metadata: \"s3://nextstrain-data/files/workflows/<pathogen>/metadata.tsv.zst\"\n sequences: \"s3://nextstrain-data/files/workflows/<pathogen>/sequences.fasta.zst\"\n\nadditional_inputs:\n - name: private\n metadata: \"data/private_metadata.tsv\"\n sequences: \"data/private_sequences.fasta\"\nIf you would like to run the phylogenetic workflow without the Nextstrain inputs,\nthen you can use the inputs parameter to completely override them.
inputs:\n - name: private\n metadata: \"data/private_metadata.tsv\"\n sequences: \"data/private_sequences.fasta\"\nWe will be updating our existing pathogen workflows to use the standardized\nparameters and you can track our progress in our public tracking issue.\nPathogen workflows that currently support multiple inputs with the standard\nparameters are avian influenza, West Nile virus, and zika.\nThe input configuration for the SARS-CoV-2 workflow already supports multiple\ninputs with additional features, so we will not be updating it to conform to the\nnew standard.
\nIf you have questions or comments, please feel free to post to our discussion forum\nor create an issue in a pathogen GitHub repository.
\n", "url": "https://nextstrain.org/blog/2025-09-29-standardized-multiple-inputs", "title": "Standardized Multiple Inputs", "date_modified": "2025-09-29T00:00:00.000Z", "author": { "name": "The Nextstrain team" } }, { "id": "https://nextstrain.org/blog/2025-09-24-mumps-resources", "content_html": "Expanding the collection of core pathogen datasets, we now provide regularly updated phylogenetic monitoring of mumps virus at:
\n\nThese phylogenies are generated using genomic data from NCBI GenBank, and are updated daily when new sequences are uploaded to NCBI.
\nMumps orthorubulavirus causes mumps, a viral disease spread primarily among humans that is highly contagious and spreads through respiratory droplets. The virus typically infects the salivary glands, causing characteristic swelling and inflammation, making it difficult to speak or move the jaw. The occurrence of mumps infections has been reduced by the widespread use of the MMR (measles, mumps, rubella) vaccine.
\nWe provide two views of mumps for genomic analysis, a Global and a North-America filtered dataset.
\n![\"Figure](\"/blog/img/mumps-nextstrain-dataset.png\")
\nFigure 1. Phylogenetic views of mumps viral genetic diversity. Views are split out into \"global\" and \"north-america\" focused phylogenies. Available at nextstrain.org/mumps/global and nextstrain.org/mumps/north-america.
![\"Figure](\"/blog/img/mumps-genotype-colors.png\")
\nFigure 2. Annotated by Mumps genotypes. Samples are annotated by Mumps genotypes based on both GenBank annotations and Nextclade assignments. Tree, map, and frequencies plots of the mumps genome tree are shown and available at nextstrain.org/mumps/global.
We provide two Nextclade datasets (SH region and genome) to assign genotypes (e.g. A to N) to mumps samples based on criteria outlined by the WHO and tree placement. Scaffold strains for the Nextclade dataset were pulled from the literature (Jin et al., 2005, WHO 2012, Jin et al., 2015) and augmented with some subsampling to fill in the tree. Mutations are called against a Jeryl-Lynn reference (D90232 for the SH region and HQ416907 for the full genome).
\n| Scope | \nNextclade dataset | \nReference | \n
|---|---|---|
| SH region (~316 bp region) | \nmumps/sh | \nD90232 (Jeryl-Lynn) | \n
| Full genome | \nmumps/genome | \nHQ416907 (Jeryl-Lynn) | \n
![\"Figure](\"/blog/img/mumps-nextclade-dataset.png\")
\nFigure 3. Nextclade datasets of mumps virus based on the ~316 SH region and the full genome. Scaffold phylogenetic trees were created for a Nextclade dataset and colored by genotype assignment. Below the trees, the SH region is highlighted in a magenta box to highlight the difference between the SH and full genome datasets. The Nextclade datasets are available as mumps/sh and mumps/genome.
We curate sequence data and metadata from NCBI as the starting point\nfor our analyses. Curated sequences and metadata are available as flat\nfiles at:
\nWe welcome comments or suggestions from mumps researchers to improve these Nextstrain datasets for their use case. Special thanks for feedback from Louise Moncla for answering questions and providing some biological context.
\n", "url": "https://nextstrain.org/blog/2025-09-24-mumps-resources", "title": "New Resources for Mumps", "date_modified": "2025-09-24T00:00:00.000Z", "author": { "name": "Jennifer Chang, Richard Neher, Jover Lee, Kim Andrews" } } ] }