# Human heart LFQ investigation: feasibility outcome

The processed data were downloaded and audited, but the planned ischemic cardiomyopathy (ICM) versus non-failing comparison cannot be identified from the metadata retrieved in this session. No differential-abundance results or imputation-sensitivity results are claimed. This is the approved feasibility fallback, not a completed inferential investigation.

## Evidence and cohort problem

According to the article retrieved through PubMed, Tomin et al., *Mass Spectrometry-Based Redox and Protein Profiling of Failing Human Hearts* ([DOI](https://doi.org/10.3390/ijms22041787), [PubMed](https://pubmed.ncbi.nlm.nih.gov/33670142/)), the analyzed left-ventricular cohort comprises 15 end-stage failing male hearts (12 ICM, 3 DCM) and 10 non-failing male donor hearts not accepted for transplantation. This reference is not necessarily representative of healthy living people. The article reports 2,700 quantified proteins across 25 samples.

[PRIDE PXD021261](https://www.ebi.ac.uk/pride/archive/projects/PXD021261) provides proteinGroupss_LFQ.txt, a 16,302,570-byte processed MaxQuant table with 2,917 protein groups and 30 numeric LFQ experiment identifiers. The downloaded file SHA-1 exactly matches the archive checksum: 02149d5c49eea651694c4f2207a650aa92f452cc. summarys_LFQ.txt also matches its archive checksum.

All 150 rows of the community SDRF label disease as 'not available'; the annotation does not establish which experiments are ICM, DCM or non-failing. The MaxQuant parameter XML and summary map acquisition files to numeric experiment IDs and fractions, not diagnoses. The summary contains six fraction runs for each experiment except experiment 24, which has five. These fraction runs are not independent biological replicates.

The original submission metadata describes N=5 control, N=5 DCM and N=5 ICM tissue samples, unlike the published 25-person cohort, and mentions two measurement batches and subsequent normalization. This discrepancy requires reconciliation, not an assumption that either count maps directly to the 30 LFQ columns. The published supplemental index at https://www.mdpi.com/1422-0067/22/4/1787/s1 returned HTTP 403. Its tables were not obtained or assessed; they may contain needed mappings. No attempt was made to infer diagnosis from numeric identifiers or clustering.

## Completed missingness audit

| Metric | Observed result |
|---|---:|
| Input protein groups | 2,917 |
| Groups excluded by contaminant, reverse or site-only flags (union) | 217 |
| Retained protein groups | 2,700 |
| Unassigned LFQ experiment columns | 30 |
| Missing entries across retained groups and columns | 35.95% |
| Per-column missingness range | 26.33–50.89% |
| Groups observed in all 30 columns | 777 |
| Groups missing in all 30 columns | 0 |

Zero LFQ values and existing NA were treated as missing for this descriptive audit. Positive observed values were preserved. No imputation, disease-specific filtering, batch correction, differential test, FDR calculation or causal inference was performed. The 777 complete groups are an all-column descriptive count, not a selected-cohort analysis set. Removing incomplete groups would retain only 28.8% of the 2,700 groups across these 30 columns and would preferentially retain well-measured proteins; the valid ICM/reference subset may have different completeness.

## What the data currently support

The processed matrix is available, modest in size and reproducibly auditable. Missingness is substantial and heterogeneous across experiments. The available evidence does not identify any protein as higher or lower in the intended ICM group because that group's columns are not established. This is an inability to estimate the requested contrast, not evidence of no biological differences. Missingness alone cannot distinguish random missingness from low-abundance censoring, and does not establish how much imputation would change group effects.

## Concrete prerequisites for continuation

Obtain an authoritative mapping for each of the 30 experiment IDs to donor identity, diagnosis (ICM/DCM/non-failing), inclusion in the published 25-sample analysis and exclusion reasons for other experiments. Resolve repeated donors if any, acquisition batch membership and the normalization applied after the archived MaxQuant LFQ table. The source could be the original supplementary tables or a depositor-provided sample sheet. Do not silently pool disease etiologies or substitute another cohort.

Once this mapping is validated, the approved analysis can apply the prespecified >=70% observed-per-group primary filter, estimate ICM-minus-reference log2 effects with uncertainty and BH FDR, and compare available-case results against complete-observation and repeated low-intensity-imputation results on a controlled protein universe. Proteins unobserved in an entire group require separate detection evidence, not invented finite fold changes. Exact design and batch adjustability depend on the verified mapping.

## Reproducible deliverable

heart_lfq_audit.zip includes five downloaded source files, sample_audit.csv, observed_lfq_unassigned.csv, missingness_mask.csv, qc_summary.json, provenance.json, reproduce_audit.py and README.txt. Source URLs, SHA-1/SHA-256 checksums and Python/pandas/numpy versions are recorded. Run `python reproduce_audit.py` from the extracted directory; no network is required. The standalone script was executed, matched the QC summary and reproduced the three saved CSV tables with identical SHA-256 hashes. The ZIP integrity check passed.

The managed local notebook was used because no enabled remote hosts were returned and these processed inputs are small. No raw MS reprocessing was performed. The statistical and sensitivity phases remain blocked by unresolved cohort metadata.
