Scientific analysis workflow

Core turns heterogeneous proteomics inputs into reviewable scientific artifacts. The central operational rule is to preserve every acceptance, rejection, mapping, and policy decision between raw input and biological interpretation. A successful process exit is not evidence that every record participated in the result.

flowchart LR
    A[Source files] --> B[Detect and validate format]
    B --> C[Normalize run bundle]
    C --> D[Validate design and sample identity]
    D --> E[Identification and FDR]
    E --> F[Protein inference and quantification]
    F --> G[Statistics and interpretation]
    G --> H[Review artifacts and claims]
    B -. rejected records .-> I[Issue ledger]
    D -. blockers .-> I
    E -. ambiguous evidence .-> I
    F -. missingness .-> I
    I --> H

Establish the input contract

1. Detect the source format and run its validation surface before conversion.
2. Parse FASTA, spectra, identifications, and experimental design with the strictness appropriate to the study. Retain accepted and rejected records.
3. Build a normalized run bundle. Its manifest anchors source hashes, generated files, record counts, rejection counts, schema version, and run metadata.
4. Confirm sample identifiers, conditions, replicates, batches, pairing, and referenced files before computing a contrast.

Stop if a required file is missing, the design cannot identify the intended comparison, target-decoy labeling is ambiguous, or normalization would proceed without the required quantitative context.

Produce evidence by level

1. Normalize PSMs and record score orientation and search-engine dialect.
2. Estimate error rates at the level being claimed. Peptide, protein, and PTM site results cannot inherit a PSM-level threshold by implication.
3. Preserve peptide-to-protein ambiguity through grouping or an explicit shared-peptide policy before protein rollup.
4. Apply quantitative normalization and missingness policy before differential testing. Record condition order, covariates, pairing, and multiple-testing correction.
5. Add annotations, pathways, complexes, regulators, or drug-target context only after the statistical evidence is stable.

Review the result

Review the typed artifacts, not only charts or prose. The minimum handoff contains the normalized-run manifest, validated workflow configuration, acceptance and issue ledgers, FDR evidence, protein-inference result, quantification and statistical reports, and the biological claim bundle.

A rerun is comparable only when source hashes and policy-bearing configuration match. A changed threshold, normalization method, database, design mapping, or annotation release is a new evidence context and should be labeled as such.

Start with Entrypoints and worked examples, then use Scientific data contracts to inspect each artifact boundary.