MSU Asset Acceleration: How University Infrastructure Plugs into Validated Pipelines

Audience: MSU faculty, ICER, Genomics Core, Pharm & Tox, ADDRC
Context: Mapping existing MSU infrastructure to 🔧🦎 ecoPrimals validated pipelines
License: CC-BY-SA 4.0
Last Updated: March 17, 2026

Overview

The 🔧🦎 ecoPrimals springs are not academic exercises. They are validated scientific pipelines that reproduce published results across seven quantitative domains — built to run on consumer hardware without institutional infrastructure. That means they run faster and more reliably when institutional infrastructure is available.

This document maps each major MSU asset to the spring that directly consumes it, the current capability level, and the acceleration that institutional integration unlocks.

Asset 1: MSU Genomics Core — Illumina/Nanopore Sequencing

Current state: The Genomics Core produces sequencing data (Illumina 16S amplicon, whole-genome, Nanopore long-read). Downstream analysis typically happens in Python/R notebooks on the researcher’s laptop using QIIME2, DADA2, or custom scripts. Results are rarely reproducible because the analysis environment is not standardized and provenance is not tracked.

What 💧♨️ wetSpring provides:

Genomics Core output → wetSpring sovereign 16S pipeline
  ├── 30 pure-Rust modules (zero Python dependencies)
  ├── DADA2-equivalent denoising (validated against public BioProjects)
  ├── GPU spectral matching: 1,077× speedup over CPU Python
  ├── Full provenance chain: sample → demux → denoise → taxonomy → publication
  └── Exit 0/1 validation: results either match known ground truth or fail loudly

Validation evidence:

• 376 experiments, 5,707+ checks, all PASS
• 4 public BioProjects benchmarked (ERP022042, SRP151114, SRP199294, SRP354276)
• Python/Rust parity to ≤ 10⁻¹² on all diversity metrics
• Handles: 16S amplicons, shotgun metagenomes, cold seep deep-sea samples, agricultural soil, clinical microbiome, water treatment metagenomes

MSU integration path:

1. Genomics Core delivers FASTQ files (existing workflow, no change)
2. 💧♨️ wetSpring pipeline runs locally (researcher’s machine or ICER compute node)
3. Output: OTU table + diversity indices + Anderson disorder parameter W + full provenance
4. Results are signed, reproducible, independently verifiable

Acceleration unlock: Every Genomics Core sample processed through 💧♨️ wetSpring gets a provenance receipt. The researcher can prove, cryptographically, that the analysis ran correctly and the result is what they claim it is. This is ISO 17025 readiness without additional process overhead.

Asset 2: ICER HPC — CPU/GPU Cluster

Current state: ICER provides access to CPU clusters and NVIDIA V100/A100 GPUs. Researchers typically submit SLURM jobs running Python or compiled C++/CUDA code. The software environment is managed by module files, creating reproducibility issues (different Python versions, different library versions, different CUDA versions).

What 🔧🦎 ecoPrimals provides:

♻️📦 ecoBin standard: pure Rust, zero C dependencies, static binary. A compiled 🔧🦎 ecoPrimals spring binary:

• Has no external dependencies (no Python, no system libraries, no CUDA runtime)
• Produces the same output on ICER V100 that it produces on a consumer RTX 3090
• Compiles once, runs anywhere (any Linux, any GPU via WebGPU/Vulkan)

# Build once (developer machine)
cargo build --release --target x86_64-unknown-linux-gnu
# scp binary to ICER login node
scp target/release/wetspring [email protected]:~/

# Run on ICER GPU node (no module loads, no conda, no spack)
sbatch --gres=gpu:1 ./wetspring validate_anderson --sample SRR123456
# exit 0 = PASS, exit 1 = FAIL

Acceleration unlock: ICER’s A100s provide ~4–8× GPU compute vs consumer RTX 3090 for f64 operations (no CUDA throttle on data center GPUs). Large-scale analyses that take hours on a consumer GPU take minutes. The Anderson spectral sweep across 376 experiments could run simultaneously across all ICER nodes — 15,000+ checks in a single SLURM array job.

Specific ICER-accelerated workloads:

Reproducibility note: Because 🔧🦎 ecoPrimals uses WebGPU/Vulkan (not CUDA), results are vendor-agnostic. An analysis started on an NVIDIA consumer GPU and completed on an ICER AMD node will produce the same floating-point result. This is mathematically verifiable — the test suite enforces it.

Asset 3: ADDRC / HTS Facility — 8,000+ Compound Library

Current state: The ADDRC (Assay Development and Drug Repurposing Core, Erika Lisabeth, Director) runs high-throughput screens against the compound library using JAK/cytokine pathway assays. Data analysis happens in Excel and GREENScreen. Drug prioritization for follow-up is done by pathway analysis (MATRIX scoring) without tissue geometry.

What the drug discovery pipeline provides:

Anderson-augmented MATRIX scoring adds a spatial geometry dimension to standard pathway-based drug-disease scoring (see DRUG_DISCOVERY_PIPELINE.md for full detail). The result: a ranked compound list that accounts not only for whether a drug hits the right pathway but whether it can physically reach its target.

ADDRC compound library (8,000+ compounds)
    → Anderson-augmented MATRIX scoring (ecoPrimals nS-605)
    → Priority ranking with tissue geometry rationale
    → iPSC validation of top candidates (literature-aligned MSU Pharmacology benchmarks)
    → Medicinal chemistry optimization (Ellsworth)

Current capability:

• 6 candidates scored computationally (329/329 checks PASS)
• Scaling to 8,000 compounds requires only compound metadata (MW, delivery route, target pathway) — no additional wet lab data needed
• Runs in < 1 second on consumer GPU (< 0.1 second on A100)
• Anderson geometry scoring is open-source, inspectable, modifiable

Integration path:

1. ADDRC provides compound metadata (existing GREENScreen data) as CSV
2. 🔧🦎 ecoPrimals scoring pipeline produces ranked list with geometry rationale
3. Top candidates proceed to iPSC validation against published Pharmacology benchmarks
4. HTS data from the screen feeds back into Anderson model refinement
5. Rho/MRTF inhibitor literature (Neubig group) evaluated for AD cross-talk using same scoring

Asset 4: MSDS Program — Graduate Student Talent

Current state: MSDS students complete a capstone project, typically a Jupyter notebook ML model trained on a public dataset. Most projects do not produce reproducible results and most pipelines cannot be run by anyone outside the team.

What K-Nome provides:

K-Nome (see KNOME_TEACHING_BRIEF.md) is the methodology that produced the 🔧🦎 ecoPrimals springs. Adapted as a pedagogy:

• Students reproduce a published result from their advisor’s domain
• They build it in Rust with explicit validation checks
• The Rust compiler is the fitness function — it rejects incorrect implementations
• Output: a binary that exits 0 on all checks and documents its own provenance

What MSDS students can contribute:

🔧🦎 ecoPrimals springs validate against published methods and datasets in each domain; capstone projects reproduce a peer-reviewed result and port it to sovereign Rust.

Student outcome: A validated, reproducible, publicly documented implementation of a key paper from their advisor’s domain. Runs on any hardware. Independent of the lab’s internal data. Publishable as a software note.

Asset 5: Faculty Research Computing — Individual Lab Infrastructure

Current state: Labs maintain individual compute resources — workstations, lab servers, department clusters — that are incompatible with each other and with institutional HPC. Moving data between these environments requires manual coordination, format conversion, and environment setup.

What the ⚛️🧬 NUCLEUS bonding model provides:

⚛️🧬 NUCLEUS is the 🔧🦎 ecoPrimals deployment architecture that composes distributed hardware into a coordinated mesh. Three bonding types:

What this enables:

• A 💧♨️ wetSpring analysis started on a department workstation can dispatch heavy GPU computation to ICER overnight — automatically, without manual job submission
• Results appear on the lab workstation with full provenance
• ADDRC HTS data stays on ADDRC hardware — only the computation crosses the network, not the raw data
• No cloud upload, no FTP, no institutional data governance concerns

Timeline: ⚛️🧬 NUCLEUS deployment requires toadStool + 🐟⚡ barraCuda (both public) on each node. Installation: 30 minutes. Configuration: automatic discovery via BirdSong protocol (encrypted UDP, zero metadata leakage).

Consolidated Acceleration Map

Getting Started

All spring repositories are public and require only Rust (stable):

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
git clone https://github.com/syntheticChemistry/wetSpring
cd wetSpring/barracuda
cargo test --workspace    # 1,443+ tests, should exit 0

The ICER module for ♻️📦 ecoBin binaries is:

module load rust/stable   # or install Rust directly — 5 minutes

No CUDA, no conda, no spack, no module file conflicts.

Contact for collaboration and access: see contacts.md in this directory.
Spring repositories: github.com/syntheticChemistry/
Primal repositories: github.com/ecoPrimals/