henhaobio
Sign in

bioRxivBioinformaticsPreprint

Nextstrain automates real-time phylogenetic analysis of open data for endemic and emerging pathogens

Kimberly R. AndrewsiD, Jennifer ChangiD, Cornelius RoemeriD, James HadfieldiD, Victor LiniD, Anderson Fernandes BritoiD, Richard Olumide DaoduiD, I. A. JoiaiD, Kathryn E. KistleriD, Allison W LiiD, Louise H. MonclaiD, Miguel I. ParedesiD, Denise KühnertiD, Laura Marcela TorresiD, Laura VoitliD, Ivan AksamentoviD, Emma B. HodcroftiD, John HuddlestoniD, John T. McCroneiD, John SJ Anderson, Thomas R. SibleyiD, Jover LeeiD, Richard A. NeheriD, Trevor BedfordiD

PDFDOICodeData

In-depth summary

Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (17 Jul 2026) · CC BY 4.0

Standardized Snakemake pipelines on GitHub Actions fetch open sequence data, curate it, and publish daily or weekly updated phylogenies for 21 viruses and Mycobacterium tuberculosis.

Genomic surveillance platforms need analyses that stay current as sequences are deposited. Compared with the 2018 Nextstrain release, where Augur was a monolithic repository with per-pathogen configuration, each pathogen now has its own modular workflow with Nextclade quality control and fully automated updates.

Key results

  1. ShownNextstrain maintains 22 core pathogen analyses. Nineteen are automated real-time analyses of open data (18 viruses, one bacterium). The number of phylogenies per pipeline varies widely.

    Phylogenies per pipeline range from one (Zika, rabies, M. tuberculosis) to 87 (SARS-CoV-2); most pipelines produce two to ten. Ebola and enterovirus D68 are not fully automated; seasonal influenza uses restricted data only. · 3.1 Summary of pipeline features; Tables

  2. ShownViral analyses use two main Snakemake workflows (ingest and phylogenetic), plus an on-demand nextclade workflow for some viruses. A GitHub Actions job runs the costly phylogenetic step only when new sequences are detected. Large pipelines typically run weekly.

    Typical subsampling gives about 3000-5000 sequences; the workflow aligns them to a reference, builds a tree with IQ-TREE (GTR) and refines it with TreeTime. Most pipelines run daily; SARS-CoV-2, influenza, mpox and RSV run weekly on AWS Batch. · 2.2 Viral analysis pipeline; 2.2.2 Phylogenetic workflow; 2.2.3 Automation

  3. SuggestedThe authors report that a few thousand viral sequences is a practical working size. Tree building becomes the bottleneck at larger sizes.

    A few thousand sequences run in roughly 1 hour on a laptop, and the Auspice visualization loads in under 0.5 seconds. Beyond 10k or 20k tips, CPU time and memory become limiting. The supporting scaling table is for SARS-CoV-2 and is not reproduced in the text. · 2.2.2 Phylogenetic workflow; Tables

  4. ShownThe M. tuberculosis pipeline starts from raw Illumina reads rather than consensus genomes. It subsamples about 1000 samples per run and caches per-sample results to limit compute, and it also predicts drug resistance with TBProfiler.

    About 1000 samples are chosen per run from roughly 170,000 SRA samples with Illumina shotgun data. Snippy produces the alignment, which is converted to a compact VCF for IQ-TREE. TBProfiler predicts drug resistance and lineage. Snippy and TBProfiler results are cached in S3, and the workflow runs weekly. · 2.3.1 Ingest and phylogenetic workflow; 2.3.2 Automation

  5. SuggestedNextstrain responded quickly to outbreaks. For mpox it developed a new real-time pipeline in 2022. For H5N1 in US dairy cattle it expanded an existing pipeline and added consensus genomes from the Andersen Lab to work around delayed GenBank assemblies.

    Avian influenza assemblies lagged SRA deposition by an average of 41 days, so the pipeline also used consensus genomes assembled by the Andersen Lab. The mpox clade IIb phylogeny (Fig. 3) shows 927 viruses; mpox needed masking of repetitive regions and adjusted alignment parameters for a genome of about 200k base pairs. · 3.2.1 Mpox; 3.2.2 Avian influenza; Figure legends

Methods, data and reproducibility

The paper describes the architecture of automated Snakemake pipelines that ingest and curate public pathogen sequences and metadata, classify them with Nextclade or TBProfiler, build time-resolved phylogenies with Augur, IQ-TREE and TreeTime, and publish results to nextstrain.org.

  • Open viral consensus genomes and metadata from GenBank and Pathoplexus for 21 viruses · virus · n = all available sequences; typically 3000-5000 per subsampled phylogeny
  • SARS-CoV-2 and influenza sequences from GISAID for a subset of restricted analyses · virus · n = all available data in GISAID
  • M. tuberculosis Illumina shotgun reads from SRA, subsampled for each run · Mycobacterium tuberculosis · n = about 1000 per run, from about 170,000 SRA samples
  • Mpox clade IIb time-resolved phylogeny (Fig. 3) · mpox virus · n = 927 viruses
Methods & tools
Snakemake, Augur, Auspice, Nextclade, IQ-TREE, TreeTime, MAFFT, Snippy, TBProfiler, NCBI Datasets, GitHub Actions, AWS Batch
Software & environment
The Nextstrain command-line tool supports Docker, Conda, ambient and Singularity/Apptainer runtimes. No pinned package versions are given for the pipelines.
Experimental conditions
A few thousand viral sequences take about 1 hour on a laptop. Scaling tests for SARS-CoV-2 used a single AWS c7a instance with 2 cores. Large pipelines (SARS-CoV-2, influenza, mpox, RSV, M. tuberculosis) run on AWS Batch.
Reproducibility gaps
GISAID-based analyses (a subset of SARS-CoV-2 and influenza pipelines) cannot have their curated data or results reshared. Ebola and enterovirus D68 pipelines are not fully automated.
Limitations (5)
  • Scaling beyond 10k or 20k tips bottlenecks at tree building in both CPU time and memory, so analyses typically use subsampled sets of a few thousand sequences. · stated by the authors
  • Extending the M. tuberculosis pipeline to other bacteria needs changes for homologous recombination, open pangenomes, and species-specific masking, lineage and resistance steps. · stated by the authors
  • Analyses can lag ongoing evolution, depending on how often new data become available. Restricted GISAID data limit resharing of curated data and results. · stated by the authors
  • This is a descriptive software and resource paper. It reports no quantitative benchmark against other surveillance platforms and no accuracy validation of the phylogenies or subsampling. · noted in this summary
  • The outbreak examples (mpox, H5N1) are summarized narratively with little supporting statistics in the text. Performance figures come from a single SARS-CoV-2 scaling table that is not reproduced here. · noted in this summary

Read this if you want to run, adapt or reuse Nextstrain's automated pathogen phylogenetics workflows or its open curated sequence and metadata files.

Citations
1

Abstract

MotivationGenome sequencing provides an exceptional window into the evolutionary and epidemiological dynamics of endemic and emerging pathogens, and thus allows for better, more targeted, public health interventions. Online genomic surveillance platforms can provide near real-time insight into these dynamics. ResultsNextstrain provides continually updated real-time genomic surveillance for 21 viruses and the bacterial pathogen Mycobacterium tuberculosis, with most analyses relying solely on open sequence data. Each pathogen includes steps to fetch and curate open data, classify sequences using established nomenclature systems, perform phylogenetic analyses, and share the results publicly. These analyses are automated, with most running daily to provide continually updated snapshots of pathogen evolution. Availability and ImplementationAll source code is available at https://github.com/nextstrain. Phylogenetic results can be visualized and downloaded at https://nextstrain.org/pathogens, and open sequence data and curated metadata are available at https://nextstrain.org/pathogens/files.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISHadfield '17Neher '18Szarvas '20Varilly '25Majidian '26Foster '25
75% semantic matchEvolutionary Biology
J. Hadfield, C. Megill, S. M. Bell+6 more·bioRxiv·Nov 22, 2017