# Frozen analysis specification — PXD021261 recovery
Frozen before group-level hypothesis tests on 2026-09-28.

## Cohort and provenance
Use explicit Table S5 RawValues headers: 12 ICM and 10 NF male left-ventricular samples, one sample per heart as described by the article. Exclude DCM 621/641/648 and archive-only experiments 11/24/65/115/128. Original omission reasons for those five are unknown; do not assert they are replicates. Six fraction runs per included experiment are already integrated in the MaxQuant LFQ column.
Sources: PRIDE PXD021261 and Table S5 from the official PMC Open Access mirror, byte-identical to Europe PMC. The earlier audit and review history remain preserved. Batch membership and individual age/EF/clinical covariates are unavailable; no batch or covariate adjustment can be justified from inferred labels.

## Preprocessing and primary analysis
Start from archived proteinGroupss_LFQ.txt, rather than the paper's prefiltered/imputed table. Remove the union of Potential contaminant, Reverse and Only identified by site '+' flags (217/2917 groups); retain protein-group IDs without collapsing to gene names. Convert zero LFQ and NA to missing and log2-transform positive LFQ. Primary normalization is the archived MaxLFQ normalization, with no additional centering or imputation.
Test groups observed in at least 9/12 ICM and 7/10 NF samples (1327 groups). Use difference of observed log2 group means (ICM minus NF), with unequal-variance Welch standard errors, Welch–Satterthwaite degrees of freedom, two-sided p-values and 95% t confidence intervals. This is unmoderated two-group linear-model inference: limma is unavailable; no empirical-Bayes moderation is claimed. Apply Benjamini–Hochberg across all 1327 primary tests; discovery threshold q<0.05. No additional fold-change threshold; report continuous effects and uncertainty. Confidence intervals are pointwise, not simultaneous.
Check full-rank two-group design, per-protein sample counts, residual standardized outliers, variance ratios, and leave-one-donor-out effect direction and FDR stability. Do not exclude samples based on results.

## Missing-data sensitivity
Keep the primary 1327-group universe and observed measurements fixed.
1. Complete observations: test only proteins measured in all 22 selected samples. Report BH across this subset and conservative fixed-universe BH (other primary groups assigned p=1 solely for correction, not called tested). Distinguish changes in testability from changed effects.
2. Repeated low-intensity imputation: 5 parameter pairs (downshift, width) = (1.0,0.3), (1.8,0.3), (2.5,0.3), (1.8,0.2), (1.8,0.5). Parameters multiply the per-sample observed log2 SD, using mean/SD across all 2700 retained groups; truncate the normal above the sample's observed fifth percentile. Fill missing entries only, and run the same Welch test/BH independently for 20 seeds (20260928 through 20260947) per pair. These are conditional stress-test results, not uncertainty-calibrated multiple-imputation inference: do not pool p-values or treat filled values as new biological replicates. Report median/range effects, direction agreement, q ranges and fraction of runs q<0.05.
3. Detection: two-sided Fisher exact tests for observed-versus-missing counts in ICM/NF across all 2700 groups, with a separate BH family. No fold-change claim for a protein unobserved in one entire group.
4. Normalization: repeat available-case tests after subtracting each sample's median observed log2 LFQ and adding the grand median. This assumes most proteins are stable and is an explicit sensitivity analysis, not inferred batch correction.

## Interpretation and exports
Call a primary discovery stable under the assessed missing-data scenarios only if its direction is unchanged and q<0.05 in all 100 imputation runs; report complete-case and normalization support separately. Report primary findings that fail these criteria as assumption-sensitive. Do not equate no discovery with absence of a biological difference.
Export all 2700 annotated groups with testability flags, full primary and sensitivity tables, sample QC, observed matrix, missingness mask, figures, script, software versions, source checksums and this specification. General-purpose report figures; no journal-compliance claim.
Conclusions are exploratory associations in end-stage male bulk tissue. Donor selection, medications, tissue composition, unknown batches and missingness mechanism limit interpretation. No raw-spectrum reprocessing or causal inference.
