BV-BRC Planetary Grand Challenge: Streaming Sieve & Multi-Clock Deconvolution of 10,000–50,000 Genomes
Overcoming the 10,000-taxon Bayesian MCMC scaling barrier across 58 years of Influenza A/H3N2 surveillance (1968–2026) in under 50 seconds on commodity hardware.
While conventional phylogenetic pipelines become computationally demanding on very large datasets, ChronAeon completes live streaming ingestion, high-throughput sieve triage, outlier quarantine, and multi-clock community deconvolution in 49.58 seconds total wall-clock time across 10,000 isolates. With zero metadata priors, ChronAeon autonomously isolates the accelerated wild waterfowl reservoir (Community 4: 99.0% avian, $2.7\times$ faster clock), the North American swine reservoir (Community 2: 98.9% swine), and the post-lockdown human resurgence (Community 0: 100% human).
Interactive Multi-Host Reservoir & Rate Spectrum Explorer
Deconvolved clock regimes ($K^*=7$) across 9,977 analysis-ready genomes. Click any community below to inspect host proportions, evolutionary rates ($\mu$), and $t_\mathrm{MRCA}$ horizons.
Comparative Scaling: Bayesian Tree Search Suites vs. ChronAeon
| Evaluation Dimension | Standard BEAST 2 / BEAST X | TargetedBeast (Bouckaert 2025) | Parallel SCA (Shao et al. 2026) | ChronAeon Planetary Engine |
|---|---|---|---|---|
| Maximum Taxon Scale | $N \le 500\text{--}1{,}000$ | $N = 2{,}000\text{--}10{,}000$ | $N = 190\text{--}1{,}000$ | $N = 10{,}000\text{--}50{,}000+$ |
| 10,000-Taxon Execution Time | Intractable / MCMC Stagnates | 80 core-days (1,920 CPU hrs) | Memory overflow on large $S$ demes | 49.58 seconds total |
| Hardware Requirements | Dedicated HPC cluster node | HPC cluster node | 16–32 core cluster node | Standard Commodity Laptop |
| Tree Topology Requirement | Sampled via MCMC moves | Permanently frozen (fixed tree) | Sampled via structured coalescent | Tree-Free Continuous Manifold |
| Molecular Clock Model | Single global strict/UCLD clock | Fixed global strict clock | Multi-deme structured clock | Unsupervised Multi-Clock ($K^*=7$) |
| Multi-Host Reservoir Resolution | Requires discrete trait prior XML | Unmodelled (single host assumed) | Requires pre-specified deme priors | Autonomous Spectral Graph Bisection |
| Streaming Data Sieving | None (manual curation required) | None | None | Integrated (414 seq/s throughput) |
Publication Manuscript Figures: 10,000-Taxon Challenge & 50,000-Genome Atlas
Figure 1: 4-Panel 10,000-Taxon Planetary Grand Challenge (1968–2026).
(A) Computational feasibility comparing BEAST 2, TargetedBeast, Parallel SCA, and ChronAeon. (B) Streaming Sieve Triage classifying 10,000 candidate sequences in 24.1s (414 seq/s) and quarantining severe anomalies. (C) Unsupervised host-reservoir separation across the $K^*=7$ deconvolved clock communities. (D) Calibrated evolutionary rate spectrum ($\mu$) resolving wild waterfowl acceleration ($2.7\times$).
Figure 2: 50,000-Genome Pan-Pathogen Evolutionary Atlas.
(A) Continuous spectral manifold landscape (Nyström normalized Laplacian embedding). (B) Multi-clock horizons and stem-to-crown latency waterfall. (C) Evolutionary velocity spectrum across 72 leaf communities. (D) Automated surveillance firewall and anomaly typology.
Biological Narrative: Overcoming the 10,000-Taxon Frontier & Reservoir Dynamics
1. Scaling Challenges in Large-Scale Phylodynamics
In modern viral surveillance, pathogen datasets are accumulating rapidly. To investigate multi-decadal transmission dynamics, cross-species spillover events, and lineage shifts across the globe, researchers examine deep cohorts containing thousands of isolates. However, Bayesian MCMC over tree space scales super-linearly with taxon count $N$, frequently requiring extensive compute or fixed tree approximations for large cohorts. ChronAeon operates directly on sequence distance manifolds, enabling streaming triage and multi-rate deconvolution across 10,000 genomes in under a minute.
2. The Failure of Single-Clock Assumptions Across Decades & Hosts
When standard root-to-tip regression or a strict molecular clock is applied to the 58-year H3N2 cohort, the model breaks down entirely ($R^2 \le 0.10$). This breakdown does not reflect poor data quality; rather, it reveals that a multi-host pathogen population is not governed by a single molecular clock. Lineages circulating in wild aquatic birds replicate and transmit under vastly different ecological constraints than human seasonal viruses or swine enzootic strains.
3. Autonomous Reservoir Isolation Without Metadata Priors
ChronAeon resolved this multi-host heterogeneity in 23.4 seconds without ever seeing the "host" column in the BV-BRC database:
- Wild Waterfowl Avian Reservoir (Community 4, N=104): 99.0% Avian isolates. Exhibits a dramatically accelerated substitution rate of $\mu = 7.46 \times 10^{-3}$ subs/site/year ($2.7\times$ faster than the human seasonal trunk; $R^2 = 0.779$). AutoClock dated the ancestral radiation of this aquatic bird lineage to 1990.59 CE [1988.00, 1992.70].
- North American Swine Reservoir (Community 2, N=823): 98.9% Swine isolates. Operates at $\mu = 3.17 \times 10^{-3}$ subs/site/year ($R^2 = 0.687$) with an introduction date of 1999.81 CE [1999.13, 2000.44], pinpointing the triple-reassortant swine emergence that later contributed to the 2009 pandemic.
- Human Seasonal Trunks & Resurgences (Communities 0, 1, 3, 5): High-fidelity separation of the 1968 ancestral pandemic founder lineage (Community 3, restricted cubic spline deceleration 0.74x to 1965.90 CE), the Clade 3 decadal trunk (Community 1, $\mu = 2.75 \times 10^{-3}$, $t_\mathrm{MRCA} = 2010.11$), and the modern post-lockdown resurgence (Community 0, 100% human, $\mu = 4.14 \times 10^{-3}$, $t_\mathrm{MRCA} = 2019.87$).
4. The Sieve Firewall: Protecting Phylogenetic Engines at 414 seq/s
In large public databases, uncurated feeds contain severe sequencing errors that catastrophically mislead phylogenetic reconstruction. In 24.1 seconds, ChronAeon's streaming sieve (`chronaeon triage`) screened all 10,000 candidate genomes and quarantined 14–23 severe anomalies:
- 9 Non-Target Contaminants: Divergence $D > 0.78$ subs/site against nearest temporal anchors (misannotated non-H3 segments such as H1/H5 HA or NA genes).
- 3–12 Chimeric / Recombinant Constructs: Temporal discrepancies $\Delta t = 15.5\text{--}52.4\text{ years}$ between 5' and 3' halves (in vitro chimeric constructs or assembly errors).
- 2 Degraded Sequences: Excess ambiguous bases ($>15\%\text{--}20\%$ Ns).
Autonomous Reproducibility Protocol & Execution Guide
To reproduce the BV-BRC live streaming ingestion, sieve triage, and AutoClock deconvolution from scratch, use the following verified commands:
Option A: Test Live BV-BRC REST API Streaming
# Stream recent real-time 2024-2026 records directly from BV-BRC REST API
pip install -e chronaeon/ # or: export PYTHONPATH="src:$PYTHONPATH"
cd bvbrc_h3n2_sieve_grand_challenge
python3 stream_bvbrc_surveillance.py --test-stream --limit 10Option B: Run the 10,000-Taxon Sieve & AutoClock Pipeline
# 1. High-Throughput Streaming Sieve Triage (24.1 seconds, 414 seq/s)
python3 -m chronaeon.cli triage \
-a data/h3n2_challenge_10k.fasta \
-d data/h3n2_challenge_10k_metadata.csv \
-s data/h3n2_challenge_10k.fasta \
--stream-dates data/h3n2_challenge_10k_metadata.csv \
--date-col decimal_date \
--strain-col genome_id \
--n-anchor 220 \
--n-bins 32 \
-o results/h3n2_10k_sieve_report.csv \
--clean-out results/h3n2_10k_sieved_clean.fasta \
--sus-out results/h3n2_10k_sieved_sus.fasta
# 2. Unsupervised Multi-Clock AutoClock Deconvolution (23.4 seconds)
python3 -m chronaeon.cli autoclock \
-a results/h3n2_10k_sieved_clean.fasta \
-d results/h3n2_10k_sieved_clean_metadata.csv \
--date-col decimal_date \
--strain-col genome_id \
-k 8 \
--output-dir results/clock_communities \
-o results/h3n2_10k_autoclock_results.json \
-c results/h3n2_10k_autoclock_classified.csv
# 3. Generate Publication Figures
python3 generate_challenge_figure.py