Single-file, browser-only FASTA intake. NJ quick tree, sampled-pair ML screen, SPR-MCMC topology scouting, posterior clade support, averaged branch lengths, tree microscope, matrix export, and dataset nose. No backend. No CDN. No external calls. Nothing leaves your machine.
primitive HTMLlocal FASTA onlyup to 500 genomesNJ + sampled-pair ML + SPR-MCMCtree microscopeno external callsseed 1872500-file pilelight mode
1 filesave it, open it, run it locally
0 callsno frameworks, no CDN, no upload
3 pathsNJ scout, ML screen, MCMC
1. Load genomes
Drop FASTA files here
Accepts .fasta .fa .fna .fas .txt Up to 500 files per drop. Repeat to build pile.
Files are read locally. Nothing is uploaded.
No genomes loaded. Drop or choose FASTA files above.
Genome pile
#
Name
Bases
GC%
IUPAC
Burden
No genomes loaded.
Names cleaned for Newick. Duplicate file+header combinations skipped. Run NJ first to check pile before ML or MCMC.
2. Build tree
Honest label — exploratory primitive. ML: deterministic sampled-pair distance likelihood hill-climbing with SPR topology and branch-length moves. MCMC: SPR topology moves with Metropolis-Hastings acceptance, accumulating posterior clade frequencies and averaged branch lengths. Large piles score a sampled subset of pairwise distances (Score pairs cap). Fast structure scouting. IUPAC ambiguity codes are accepted; RNA U is normalized to T; circular/viroid mode wraps k-mers across the FASTA cut point. Ambiguous symbols are reported as missing data and never counted as exact k-mers. Tamura-Nei 93 is available as an aligned/position-aware distance option, useful when records are homologous and comparable site-by-site. Not IQ-TREE, MrBayes, BEAST, RAxML, or MAFFT+Geneious. Fit sharpness (σ): this is a penalty sharpness parameter on the patristic distance regression, not a substitution-model sigma; it controls how steeply the scoring penalizes residuals after OLS rescaling of predicted to observed distances. SPR acceptance rate: a fraction of SPR proposals are topologically trivial no-ops (regraft at the original parent); the displayed acceptance rate is therefore a slight overestimate of true mixing efficiency. Tamura-Nei 93 base frequencies are computed from per-genome whole-sequence counts and averaged per pair, not solely from the pairwise aligned columns.
Load at least three genomes to enable tree building.
* SSE is computed after OLS rescaling of predicted patristic distances to observed distances. Not raw residual SSE. See Download Run JSON for full method notes.
Dataset nose
Run a tree or compute distances to smell the pile.
No outlier report yet.
Method agreement + stability checks
Run NJ, ML, and/or MCMC to compare split agreement.
Shuffle and jackknife checks appear here.
Tree microscope
Run a tree. Nodes are clade-colored, branch order gets a soft gradient, and hot branch colors mean edge stress. Click a node to freeze rotation and show accession. Hover still gives genome detail.