Chapter 10: Results — wetSpring

Life science and analytical chemistry: sovereign 16S pipeline, quorum sensing, phylogenetics, PFAS — 1,368 checks across 56 experiments.

📐 Architecture-ready

10.1 Validation Summary

wetSpring is the largest spring by experiment count and validation checks: 56 experiments, 1,368 checks (1,168 CPU + 200 GPU), all passing. It validates BarraCuda and the ecoPrimals infrastructure against 16S metagenomics, quorum sensing models, phylogenetic inference, PFAS analytical chemistry, deep-sea metagenomics, and enzyme evolution — spanning six faculty connections across three institutions.

Table 10.1 — Phase Summary

PhaseDomainExperimentsChecksStatus
1–2Galaxy/QIIME2 16S bootstrap492PASS
1–2asari LC-MS + PFAS screening426PASS
3GPU diversity + spectral matching238PASS
4Sovereign 16S pipeline (end-to-end)137PASS
5Algae pond + VOC peak validation256PASS
6Public data benchmarks (4 BioProjects)1202PASS
Waters lab QS/c-di-GMP models6100PASS
Liu lab phylogenetics8137PASS
Jones lab PFAS + spectral349PASS
Anderson deep-sea metagenomics6133PASS
BarraCuda CPU + GPU parity5182PASS
GPUGPU pipeline validation8200PASS
MiscFaculty proxies, alignment, Felsenstein6116PASS
Total561,368All pass

10.2 Sovereign 16S Pipeline

wetSpring’s headline result: a complete 16S metagenomics pipeline in Pure Rust + BarraCuda GPU, replacing the Galaxy/QIIME2/DADA2 Python stack.

Table 10.2 — Pipeline Module Inventory

ModuleFunctionCPU ChecksGPU Checks
FASTQ parsing + QCSequence ingestion
Adapter trimmingQuality control
DereplicationUnique sequence identification
Chimera detectionArtifact removal
OTU clusteringTaxonomic binning
Shannon/Simpson diversityAlpha diversity
Spectral cosine matchingChemical ID
Bray-Curtis dissimilarityBeta diversity
Phylogenetic compositionTree-aware analysis
HMM batch forwardProfile HMM scanning

Table 10.3 — BioProject Benchmark (Exp014: 202/202 checks)

BioProjectSamplesReference ToolMatch StatusChecks
PRJNA48817010+QIIME2/DADA2Full parity~50
PRJNA38232210+QIIME2/DADA2Full parity~50
PRJNA119597810+QIIME2/DADA2Full parity~50
Additional10+QIIME2/DADA2Full parity~52
Total202

10.3 GPU Performance

Table 10.4 — GPU Speedups

WorkloadCPU TimeGPU TimeSpeedupParity
Spectral cosine (2,048 spectra)926×≤ 1e-10
Full 16S pipeline (10 samples)2.45×88/88
Shannon/Simpson diversity15–25×≤ 1e-6
Bifurcation eigenvalues (5×5)bit-exact2.67e-16 rel
ODE parameter sweep (64 batches)abs < 0.15

The 926× spectral cosine speedup demonstrates that GPU promotion of the right kernel can transform a bottleneck into a trivial operation. The 2.45× full-pipeline speedup is modest because most 16S pipeline time is I/O-bound (FASTQ parsing), not compute-bound.


10.4 Waters Lab — Quorum Sensing Models (100 checks)

ExperimentModelChecksKey MetricStatus
Waters 2008 QS/c-di-GMP ODELasR/LasI + c-di-GMP coupled16ODE convergencePASS
Massie 2012 Gillespie SSAStochastic QS switching13Mean switching timePASS
Fernandez 2020 bistable switchHysteresis in QS circuit14Switch range > 0.3PASS
Srivastava 2011 multi-signalTwo-input Hill AND gate19AND logic correctPASS
Bruger & Waters 2018 cooperationPublic goods + cheater dynamics20Variance < 0.05PASS
Mhatre 2020 phenotypic capacitorBistability + noise exploitation18Hill ODE stabilityPASS

10.5 Liu Lab — Phylogenetics (137 checks)

ExperimentMethodChecksKey MetricStatus
Liu 2014 HMM primitivesForward/backward/Viterbi21Numerical parityPASS
Robinson-Foulds validationTree distance metric23Exact RF distancesPASS
PhyNetPy RF distancesGene tree comparison15Match PhyNetPy outputPASS
PhyloNet-HMM discordanceIntrogression detection10Viterbi accuracy > chancePASS
SATé pipelineDivide-and-conquer alignment17Alignment scorePASS
Neighbor-joining (SATé core)Distance-based tree building16Topology matchPASS
Felsenstein pruning likelihoodMaximum likelihood on tree16Likelihood matchPASS
Smith-Waterman alignmentLocal alignment15Optimal scorePASS
Wang 2021 RAWR bootstrapGene tree resampling11Bootstrap supportPASS
Alamin & Liu 2024 placementPhylogenetic placement12Placement accuracyPASS
Zheng 2023 DTL reconciliationDuplication-transfer-loss14Event countsPASS

10.6 Jones Lab — PFAS & Mass Spectrometry (49 checks)

ExperimentDomainChecksKey MetricStatus
PFAS library (Zenodo)Reference spectra26Library matchPASS
EPA PFAS MLDecision tree classification14RF F1=0.978, GBM F1=0.992PASS
MassBank spectral matchingCosine similarity9Spectral IDPASS

10.7 Anderson — Deep-Sea Metagenomics (133 checks)

ExperimentPaperChecksKey MetricStatus
Rare biosphereAnderson, Sogin, Baross 201535Rare taxon detectionPASS
Viral metagenomicsAnderson et al. 201422Viral contig assemblyPASS
Sulfur phylogenomicsMateos, Anderson et al. 202315Tree reconciliationPASS
Phosphorus phylogenomicsBoden, Anderson et al. 202413Enzyme evolutionPASS
Population genomicsAnderson et al. 201724FST, isolation-by-distancePASS
PangenomicsMoulana, Anderson et al. 202024Gene gain/loss dynamicsPASS

10.8 GPU Validation Binaries

BinaryChecksStatus
validate_diversity_gpu38PASS
validate_16s_pipeline_gpu88PASS
validate_barracuda_gpu_v314PASS
validate_toadstool_bio14PASS
validate_gpu_phylo_compose15PASS
validate_gpu_hmm_forward13PASS
benchmark_phylo_hmm_gpu6PASS
validate_gpu_ode_sweep12PASS
GPU Total200All pass

10.9 Scholarly Reproduction Log

#Paper / PipelineTrackChecksStatus
1Galaxy/QIIME2 16S (4 experiments)192PASS
2asari LC-MS (2 experiments)226PASS
3FindPFAS screening217PASS
4Public data (4 BioProjects)1202PASS
5Waters 2008 + 5 downstream papers1100PASS
6Liu 2014 + 10 phylogenetics papers1b137PASS
7Jones PFAS + spectral (3 experiments)249PASS
8Anderson 2014–2024 (6 papers)1c133PASS
9Cahill + Smallwood proxies126PASS

10.10 Connection to Constrained Evolution Thesis

wetSpring provides the strongest single piece of evidence for the constrained evolution methodology. 1,368 checks across 56 experiments and 6 faculty connections — metagenomics, quorum sensing ODEs, phylogenetic inference, mass spectrometry, enzyme evolution — all validated by the same BarraCuda kernels evolved under type-theoretic constraint.

The 926× GPU speedup for spectral cosine matching demonstrates that the constrained evolution methodology not only produces correct results but produces them efficiently. The GPU kernel was evolved under the WGSL constraint, not hand-tuned for mass spectrometry — yet it outperforms CPU by nearly three orders of magnitude.

The Anderson deep-sea experiments (133 checks) close a conceptual loop: the computational tools validated by wetSpring are the same tools proposed for analyzing LTEE frozen fossils in Biological Validation. wetSpring proves the tools work; Biological Validation proposes using them for biological validation of the constrained evolution thesis itself.


See also: