PrimAeon and the HyphAeon ecosystem
PrimAeon is a browser app for evolutionary analysis: you paste in a sequence alignment and get a selection report back in seconds. Everything runs client-side — the HyphAeon models ship as ONNX graphs executed under ONNX Runtime's WebAssembly backend, with the analysis pipeline ported to JavaScript — so your sequences never leave your device. There is nothing to install and no account to create.
Underneath PrimAeon sits an ecosystem of tools built on the same foundation model. This page explains what each piece does, what inputs it needs, and how to choose between them — including when you should use classic HyPhy instead.
The pieces
| Tool | What it is | Where it runs |
|---|---|---|
| PrimAeon | Browser app: one alignment in, one report out — selection, co-selection, DMS, phenotype; /time adds clock + temporal analyses | primaeon.org, entirely in your browser |
| HyphAeon | Python package over the foundation model — the "microscope": selection, epistasis, mutational scanning, phenotype association | pip install hyphaeon (GitHub) |
| ChronAeon | Python package over the same model — the "radar": molecular clock dating, phylodynamics, phylogeography, genomic triage | pip install chronaeon (docs) |
| HyPhy / Datamonkey | The classic maximum-likelihood methods (MEME, BUSTED, FEL, aBSREL, …); HyphAeon is a neural surrogate for MEME | datamonkey.org, v3.datamonkey.org (in-browser via WASM), or HyPhy locally |
Which tool for which question
| Question | Tool | Input required |
|---|---|---|
| Which sites in my gene are under episodic positive selection? | PrimAeon, or hyphaeon meme | Codon alignment (FASTA); tree optional |
| Which sites co-evolve? Where are the epistatic sectors? | hyphaeon epistasis | Codon alignment; tree optional |
| What is the predicted effect of every possible mutation? | hyphaeon dms | Codon alignment; tree optional |
| Which sites are associated with a phenotype or trait? | hyphaeon phenotype | Codon alignment + trait (metadata file, preset, or foreground list); tree optional |
| When did this clade or outbreak originate? | PrimAeon /time, or chronaeon date | Alignment with sampling dates (in headers or a metadata file); tree optional |
| How fast is the outbreak growing? What is R₀ / Rₜ? | chronaeon dynamics | Dated alignment or tree |
| Where did the pathogen spread, and along which routes? | chronaeon phylogeo | Alignment + location metadata; tree optional (auto-built) |
| Screen many genomes for anomalies or emerging clades | chronaeon triage / chronaeon sketch | Reference alignment + streaming FASTA (triage); unaligned FASTA (sketch) |
| I need the exact likelihood methods for a publication | HyPhy via Datamonkey | Codon alignment + tree |
A few rules of thumb:
- Start in PrimAeon when you have an alignment and want an answer fast, or when the data cannot leave your machine.
- Move to the Python packages when you need batch runs, the full command surface (spectral splits, temporal dynamics, disease prediction, alignment filtering), or GPU acceleration.
- Use HyPhy/Datamonkey when you need the statistical guarantees of the
original maximum-likelihood methods. PrimAeon reports are neural surrogate
results — every report is flagged
is_surrogateand includes a path to run the exact analysis on Datamonkey. If privacy is the concern rather than speed, v3.datamonkey.org runs real HyPhy in the browser via WebAssembly too.
Inputs
The input that matters most is the alignment:
- Codon-aware alignment (PrimAeon,
hyphaeon). Selection analyses need a nucleotide alignment of coding sequences that preserves codon frame: the alignment length must be a multiple of 3 and free of in-frame stop codons — PrimAeon validates both on upload. Produce one with a codon-aware aligner (MACSE, bealign, cawlign) or by aligning protein sequences and back-translating to codons. A plain nucleotide alignment from MAFFT or MUSCLE can introduce frameshifting gaps and will fail validation. - Alignments for ChronAeon.
chronaeon dateandautoclockalso take in-frame codon alignments (FASTA/NEXUS, or BEAST XML with tip dates);sketchworks on raw, unaligned sequences, whiletriagestreams unaligned FASTA against a reference alignment.dateanddynamicsneed sampling dates — auto-extracted from sequence headers or supplied as a metadata CSV/TSV (or Auspice JSON);phylogeoadditionally needs location metadata. - Tree (optional, Newick). Everywhere a tree is accepted it is optional:
HyphAeon and
chronaeon datefall back to a tree-free path from pairwise TN93 distances (--use-tn93), andchronaeon phylogeoauto-builds one with FastTree. Trees can also be embedded in a NEXUS alignment. - Trait metadata (
hyphaeon phenotypeonly). The trait can come from a CSV/TSV metadata file, a curated preset (--preset), or an inline foreground/background taxon list (--foreground/--background). - Making a codon-aware alignment.
chronaeon alignthreads unaligned CDS sequences onto a reference in codon frame;hyphaeon filter(ormeme --filter) can then surgically mask alignment artifacts before inference.
Privacy
PrimAeon, v3.datamonkey.org, and anything you run locally — the Python packages or HyPhy itself — never send your data anywhere. The classic datamonkey.org service runs on a server, like Galaxy.
Links
- primaeon.org — the browser app
- HyphAeon on GitHub — monorepo:
aeon-core,hyphaeon,chronaeon - HyphAeon interactive overview — methods, benchmarks, and the story behind the model
- ChronAeon docs — dating, phylodynamics, and surveillance guides
- bioRxiv preprint — the HyphAeon paper
- Model weights on Hugging Face
- hyphy.org and HyPhy on GitHub — the likelihood-method reference
- datamonkey.org / v3.datamonkey.org — HyPhy as a web service / in-browser
- hyphaeon Galaxy wrappers — installed on usegalaxy.eu for web-based batch runs;
chronaeonwrappers are in progress
Support
HyphAeon is supported in part by BRC Analytics and the NIH/NIAID.