Pipeline Overview

CETSAx is orchestrated entirely by Snakemake. The diagram below shows the complete dependency graph — from raw CSV input to final outputs — as defined in the Snakefile.

Pipeline Flow Diagram

The following diagram shows the complete orchestration of the cetsax pipeline, from raw CSV input to final outputs, as managed by Snakemake.

flowchart TD
    A[("config.yaml\n(CSV path, params)")] --> B[/data/nadph.csv/]
    B --> C[Rule: fit_curves\nEC50, Hill, Δmax, R²]
    C --> D[Rule: hit_calling\nClassify responders]
    C --> E[Rule: annotate\nFetch sequences → FASTA]
    D --> E
    C --> F[Rule: system_analysis\nPCA, clustering, enrichment]
    D --> F
    E --> F
    F --> G[Rule: train_model\nESM-2 embedding + head\n🐳 GPU rule]
    C --> G
    E --> G
    G --> H[Rule: predict\nESM-2 inference + saliency\n🐳 GPU rule]
    E --> H
    H --> I[Rule: visualize\nPlots 1–14]
    G --> I
    C --> I
    E --> I
    I --> J[Rule: network_analysis\nCo-stabilization graph]
    B --> J
    J --> K[Rule: curve_ml\nK-means curve typing]
    B --> K
    K --> L[Rule: bayesian_fit\nMCMC EC50 posteriors]
    B --> L
    D --> L
    L --> M[Rule: detailed_plots\nPer-protein curves + GoF]
    B --> M
    C --> M
    D --> M
    G --> N[Rule: model_performance\nClassification report]
    H --> N
    I --> O[("results/plots/\nModel evaluation figures")]
    J --> P[("results/network/\nGEXF graph + modules")]
    K --> Q[("results/curve_ml/\nClusters + outliers")]
    L --> R[("results/bayesian/\nEC50 posteriors")]
    M --> S[("results/detailed_plots/\nPer-protein curves")]
    N --> T[("results/model_performance_report.txt")]
    O & P & Q & R & S & T --> U[("report.html\nIntegrated Results Overview")]

    style G fill:#dbeafe,stroke:#2563eb
    style H fill:#dbeafe,stroke:#2563eb
    style U fill:#dcfce7,stroke:#16a34a

Running the pipeline

Run locally:

snakemake --profile workflow/profiles/local

Run on SLURM HPC:

snakemake --profile workflow/profiles/slurm

report.html

After a complete run, report.html at the project root provides a self-contained interactive overview of all pipeline outputs — the recommended starting point for inspection.

Rule Summary

Rule	Inputs	Key Outputs	Resources
fit_curves	data/nadph.csv	ec50_fits.csv	CPU
hit_calling	ec50_fits.csv	cetsa_hits_ranked.csv	CPU
annotate	fits + hits	protein_sequences.fasta	CPU
system_analysis	fits + hits + annotations	PCA, clustering, enrichment CSVs	CPU
train_model	fits + FASTA + cluster labels	nadph_seq_head.pt, embedding cache	GPU
predict	FASTA + checkpoint	predictions_nadph_seq.csv + saliency	GPU
visualize	predictions + fits + history	14 evaluation plots	CPU
network_analysis	raw CSV	network_graph.gexf, modules	CPU
curve_ml	raw CSV	curve clusters + outliers	CPU
bayesian_fit	raw CSV + hits	Bayesian EC50 posteriors	CPU
detailed_plots	raw CSV + fits + hits	per-protein curves + GoF plot	CPU
model_performance	supervised + predictions	model_performance_report.txt	CPU