← Back to Blog Post
Use ← → or Space to navigate · 60-Second Summary
Slide 1 of 6 · A New Continuous-Field Paradigm

Rethinking recombination from first principles with AI-accelerated algorithm design

Instead of porting discrete tree searches or triplet loops to faster hardware, we formulated recombination as continuous motion across a metric reference map in linear $\mathcal{O}(NKL)$ time:

$\mathcal{O}(NKL)$ Linear
Continuous Attention Trajectories

Embeds $K$ reference genomes on a metric map with a neutral center drain, tracking each sequence site-by-site via soft attention weights and an orthogonal ghost-donor axis.

Force Dipoles
Dyadic Wavelet Polishing

Parent switches create equal-and-opposite attention dipoles across breakpoints, resolved across four orders of magnitude in tract length by dyadic wavelets and Viterbi decoding.

Seconds in Browser
Rust, Python & WebAssembly

Scans 83 whole sarbecoviruses (31 kb) in 2.6s, 240 S. pneumoniae chromosomes (2.22 Mb) in 36.1s, and 379 primate MHC alleles in 1.1s.

Validated across 1,000,000 simulated genomes (0.58%–1.01% false-positive rate) and compiled to WebAssembly for zero-install browser execution.
Slide 2 of 6 · How the RhizAeon Stack Works

Attention maps, force dipoles, and dyadic polishing in four steps

1. Map + Center Drain
Landmarks & Root Sink ($w_{i,0}$)

Places $K$ reference genomes on a map with a neutral drain at the center. Fast-mutating branches flow into the center drain instead of triggering false alarms.

2. Attention Maps
Where Each Genome Points ($\mathbf{w}_i(u)$)

At each site $u$, every sequence distributes attention weights over the landmarks. Unsampled ("ghost") donors lift the sequence along an orthogonal ghost axis $\mathbf{d}_{\mathrm{ghost}}$.

3. Force Dipoles
Push From Old Parent, Pull to New

When a sequence jumps from Parent A to Parent B, Parent A's weight drops and Parent B's weight rises equally—creating a sharp force dipole ($\Delta \mathbf{w}_i$) tested against a Brownian bridge.

4. Dyadic Polishing
Wavelets, Viterbi & Scrubbing

Wavelet filters with doubling widths ($2, 4, 8, 16\dots$ sites) catch both 80 bp and 30 kb swaps. Viterbi decoding and likelihood polishing lock in exact edges, scrub tracts, and merge clades.

Slide 3 of 6 · Four Failure Modes Fixed

Rate bursts, ancestral clades, ghost donors, and three-parent mosaics

RhizAeon performance across rate heterogeneity, ancestral clade deduplication, ghost donors, and 3-parent mosaics
(A) The neutral center drain $w_{i,0}$ keeps false alarms at $\alpha = 0.05$ under strong branch rate bursts. (B) On 10-child recombinant clades, clade merging collapses shared tracts into $1.01 \pm 0.08$ ancestral events (98.8% exact) vs. $14.37$ calls in 3SEQ. (C–D) Orthogonal ghost projection finds extinct donors (89.2% F1) and three-parent mosaics (90.0% full parent recovery).
Slide 4 of 6 · 1,000,000-Genome Continuous Sweep

1.01% false-alarm rate and iterative scrubbing across 1,000,000 genomes

RhizAeon 1,000,000-genome continuous parameter sweep and multi-event sequential scrubbing
(A) Across 333,334 clonal alignments ($R=0$), the center drain holds false alarms at 1.01% (and 0.58% on 1,200 rate-burst nulls). (B) Detection sensitivity across mutation ($\theta$) and recombination ($\rho$) rates reaches 88.2% where mutations flank the breakpoint. (C) Iterative tract scrubbing peels back $R = 1\text{ to }10$ overlapping swaps with steady breakpoint accuracy.
Slide 5 of 6 · Whole-Genome Sarbecoviruses

78–83 full sarbecovirus genomes mapped in 2.6 seconds (vs. 4.5 days in GARD)

Whole-genome recombination landscape across 78-83 sarbecovirus genomes in 2.6 seconds
(A–B) Scanning all 78–83 sarbecovirus genomes ($30{,}956\text{ nt}$) in 2.6 seconds matches all 21 whole-genome GARD breakpoints from Lytras et al. (2022) (median offset 99.0 nt; 9–12 nt at Spike NTD and S1/S2 junctions). (C–D) Pinpoints the Laotian bat BANAL-20-52 / RaTG13 RBD exchange at a 60,000× speedup.
Slide 6 of 6 · Bacteria, MHC & WebAssembly

From 2.22 Mb pneumococcal chromosomes to 86 bp HLA micro-conversions

2.22 Mb in 36.1s
S. pneumoniae PMEN1

Maps 392 transfers across 240 chromosomes (~35× faster than Gubbins), finding the exact resistance swaps at pbp2x, pbp1a, pbp2b, folA, folP, and cps.

86–150 bp
379 Primate MHC Alleles (1.1s)

Dyadic wavelets catch 103 short Exon 2/3 pocket conversions, including the classic HLA-B*46:01 hybrid between B*15:01 and C*01:02.

WebAssembly
Run Locally in Any Browser

Try the live app at veg.github.io/rhizaeon-code or grab the Rust/Python package at veg/rhizaeon-code.

Read the full 4-minute post at 2026-10-24-rhizaeon.html · Preprint on bioRxiv (2026.10.19.760237).