SOURCES.md for every tracked directory, and drop the column nobody could define

Five directories had no provenance file: expression, pmhc, proteome, tcga,
thymus. Each has one now, carrying measured facts rather than a description --
row counts, schemas, and the split between what is experimental and what is
computed.

The plan recorded this as "9 of 12 directories". It is 5 of 12: benchmark/,
dump/, raw/ and src/ are gitignored and never pushed, and .gitignore already
says why for each. Absence there is deliberate, not a gap.

proteome/SOURCES.md points at the README beside it rather than restating it --
that file already carries organism, UniProt ID, taxid, sequence count and the
exact fetch form for all 19 proteomes, and two copies of that would drift.

The `#` column in DESCRIPTION.md is gone from all 28 rows. It was undocumented,
inconsistent between column count / row count / dataset count, and the table's
own note told readers not to rely on it. A column the document disclaims is
better deleted than explained.

DESCRIPTION.md

@@ -54,45 +54,48 @@ per-dataset audits (rows, positives, negatives, HLA counts, donor counts, residu
 [`neoantigens/SOURCES.md`](neoantigens/SOURCES.md) ·
 [`ligandome/SOURCES.md`](ligandome/SOURCES.md).
 
-> The `#` column below is undocumented and does not consistently mean column count, row count or
-> dataset count. Do not rely on it; the SOURCES files carry measured counts.
+Every tracked directory now carries one:
+[`expression/`](expression/SOURCES.md) · [`immunogenicity/`](immunogenicity/SOURCES.md) ·
+[`ligandome/`](ligandome/SOURCES.md) · [`neoantigens/`](neoantigens/SOURCES.md) ·
+[`pmhc/`](pmhc/SOURCES.md) · [`proteome/`](proteome/SOURCES.md) · [`tcga/`](tcga/SOURCES.md) ·
+[`thymus/`](thymus/SOURCES.md). The four directories without one — `benchmark/`, `dump/`, `raw/`,
+`src/` — are gitignored and never pushed; `.gitignore` says why for each.
 
-
-| Path | # | Dataset | One-line |
-|------|---|---------|----------|
-| `pmhc/pmhc_full.tsv.gz` | 6 | IEDB epitope–MHC | positive IEDB MHC-ligand binding records (human IMGT/HLA + mouse H-2) |
-| `pmhc/pmhc_shortlist.tsv.gz` | 6 | IEDB epitope–MHC (≥2 refs) | high-confidence pairs (≥2 references) |
-| `immunogenicity/chowell_rebuilt.tsv.gz` | — | Chowell corpus, rebuilt | 511,301 rows / 315,191 peptides, 19,866 immunogenic, 210 alleles, 8–11mers. Immunogenic vs **self** eluted ligands, rebuilt from the IEDB dump (Chowell et al., PNAS 2015, PMID 25831525). Carries a 6.9x cysteine artefact — see SOURCES.md |
-| `immunogenicity/chowell_rebuilt_hla_matched.tsv.gz` | — | …HLA-matched | 115,592 rows; negatives down-sampled per allele to 5 per positive of that allele, seed 20260817 |
-| `immunogenicity/kesmir_rebuilt.tsv.gz` | — | Kesmir/Calis corpus, rebuilt | 65,737 rows / 36,155 peptides, 22,613 immunogenic, 196 alleles. Immunogenic vs **non-self** eluted ligands (Calis et al., PLoS Comput Biol 2013, PMID 24204222). Requires `ligandome/viral_foreign_iedb.tsv.gz`; no cysteine artefact, but a 2x C-terminal Arg/Lys gradient |
-| `immunogenicity/kesmir_rebuilt_hla_matched.tsv.gz` | — | …HLA-matched | 59,941 rows, same matching rule |
-| `immunogenicity/iedb_labeled.tsv.gz` | 7 | IEDB immunogenicity — **ipred training set** | immunogenic (T-cell+, 34,362) vs non-immunogenic (MHC-ligand+ in healthy self, 756,506); `host_species` (human/mouse) is the cross-validation axis; ipred featurizes `peptide` (MJ/hydropathy/Kidera) |
-| `neoantigens/neoag_tested.tsv.gz` | 9 | Neoantigens — tested (workhorse) | tested patient-derived neoantigens, immunogenicity 0/1 (TESLA, Neopep, CEDAR, Gfeller, ITSNdb, GBM, VACCIMEL, Bjerregaard). **Human only by label — see the mouse-contamination caveat below.** |
-| `neoantigens/neoag_candidates.tsv.gz` | 9 | Neoantigens — candidates | untested neoantigen candidates (Neopep `not_tested`) |
-| `neoantigens/neoag_tested_mmu.tsv.gz` | 12 | Neoantigens — tested, **mouse MHC** | 1,977 murine tested neo-epitopes (IEDB), immunogenicity 0/1 (740 immunogenic); `mhc_species=MusMusculus` |
-| `neoantigens/neoag_tested_hsa.tsv.gz` | 12 | Neoantigens — tested, **human MHC in mouse models** | 717 rows from the same IEDB export whose restriction is HLA — HLA-transgenic mice plus human-host epitopes (KRAS G12D, CDK4 R24C, PMEL); `mhc_species=HomoSapiens` |
-| `neoantigens/nci_complete_minigenes.tsv.gz` | 9 | Neoantigens — complete NCI screen (variant) | 1,326 mutant 25-mer minigenes across 46 GI-cancer patients with CD8/CD4 T-cell screen outcomes (28 CD8-immunogenic) — a predictor-agnostic *screened* set with real negatives. Standalone variant, **minigene-level** (`peptide` = minigene, `mhc_a` blank — not epitope-deconvolved). Source: Koşaloğlu-Yalçın et al., iScience 2022, Table S6 ([doi:10.1016/j.isci.2022.103850](https://doi.org/10.1016/j.isci.2022.103850), PMID 35128348) |
-| `neoantigens/neoantigens_tested_peptides.tsv.gz` | 16 | Neoantigens — tested, **epitope-level** | 423,085 screened 8–12mers with their wild-type counterpart and restricting alleles, 178 CD8-immunogenic, 98 patients, over three datasets (`NCI` 420,786 · `HiTIDE` 1,563 · `TESLA` 736) and the authors' own `split`. This is the screen at the resolution a pMHC model is scored at — the complement to the minigene-level `nci_complete`. Harmonized from `raw/immunogenicity/NCI_dataset_only_tested.txt` by `bench/neoag/ingest_tested.py` (2026-mhcmatch-benchmark). Predicted quantities keep a `pred_` prefix; `peptide`, `peptide_wt`, `mhc_a` and `immunogenicity` are as-reported |
-| `neoantigens/nci_gartner_mmp.tsv.gz` | 45 | NCI — confirmed CD8+ minimal epitopes | 147 rows / 144 distinct minimal mutant peptides (mmp) with restricting HLA; 120 carry the wild-type mmp, so agretopicity is **measurable rather than predicted**. `split` preserves the paper's train/test. Source: Gartner et al., Nat Cancer 2021, Suppl. Tables 3 & 20 ([doi:10.1038/s43018-021-00197-6](https://doi.org/10.1038/s43018-021-00197-6)) |
-| `neoantigens/nci_gartner_nmers.tsv.gz` | 21 | NCI — full screened nmer universe | 9,541 screened mutations as WT/mutant 25-mers (144 CD8+), with gene, mutation type, RNA-seq presence and exome VAF decile — the negative universe at the unit the assay used. Same source, Suppl. Tables 11 & 12 |
-| `neoantigens/nci_gartner_samples.tsv.gz` | 15 | NCI — patient class-I genotypes | 98 patients with full HLA-A/B/C genotype, histology and per-patient screening counts. The genotype is what makes a per-patient allele-aware scan possible. Same source, Suppl. Tables 1 & 18 |
-| `tcga/tcga_neoantigens.tsv.gz` | 10 | TCGA neoantigens | 2.24M expressed mutant-peptide–HLA binders across 8,505 TCGA donors (derived from open-access TCGA somatic + expression; per-barcode HLA/neoantigens as in TCIA / Thorsson 2018) |
-| `immunogenicity/hla_pop_freqs.tsv.gz` | — | HLA population freqs | EUR/ASN/AFR class-I (A/B/C) allele frequencies (AFND) |
-| `proteome/human.fasta.gz` | 4 | Human proteome (UP000005640) | self-reference proteome for similarity search |
-| `proteome/mouse.fasta.gz` | 5 | Mouse proteome (UP000000589) | C57BL/6 self-reference proteome |
-| `proteome/*_UP*.fasta.gz` | — | 16 bacterial + viral reference proteomes | foreign-antigen references for molecular-mimicry scans and peptide-flank extraction; see `proteome/README.md` |
-| `thymus/thymus_immunopeptidome.tsv.gz` | 1 | Thymus self-peptidome (human) | HLA Ligand Atlas thymus-eluted self peptides (25.9k MHC-I + 28.0k MHC-II) — central-tolerance "self" reference for seqtree; fills the mislabeled source #1 |
-| `thymus/thymus_expression.tsv.gz` | 1 | Thymus expression (human) | thymically expressed self: HPA thymus-cluster genes (126) + atlas thymus source proteins (10.4k) |
-| `ligandome/viral_foreign_iedb.tsv.gz` | 2 | Viral ligandome (IEDB) | viral-source presented peptides (foreign reference) |
-| `ligandome/viral_orfs_gse272406.tsv.gz` | 1 | Pan-viral ORFs (GSE272406) | translated novel viral ORF proteins (foreign; *not* thymus self — spec mislabel) |
-| `ligandome/cancer_targets_tsarina.tsv.gz` | 3 | Cancer-testis antigens | curated shared tumor-antigen genes (tsarina) |
-| `summary.tsv` | — | Pipeline stats | row counts & extraction metrics |
+| Path | Dataset | One-line |
+|------|---------|----------|
+| `pmhc/pmhc_full.tsv.gz` | IEDB epitope–MHC | positive IEDB MHC-ligand binding records (human IMGT/HLA + mouse H-2) |
+| `pmhc/pmhc_shortlist.tsv.gz` | IEDB epitope–MHC (≥2 refs) | high-confidence pairs (≥2 references) |
+| `immunogenicity/chowell_rebuilt.tsv.gz` | Chowell corpus, rebuilt | 511,301 rows / 315,191 peptides, 19,866 immunogenic, 210 alleles, 8–11mers. Immunogenic vs **self** eluted ligands, rebuilt from the IEDB dump (Chowell et al., PNAS 2015, PMID 25831525). Carries a 6.9x cysteine artefact — see SOURCES.md |
+| `immunogenicity/chowell_rebuilt_hla_matched.tsv.gz` | …HLA-matched | 115,592 rows; negatives down-sampled per allele to 5 per positive of that allele, seed 20260817 |
+| `immunogenicity/kesmir_rebuilt.tsv.gz` | Kesmir/Calis corpus, rebuilt | 65,737 rows / 36,155 peptides, 22,613 immunogenic, 196 alleles. Immunogenic vs **non-self** eluted ligands (Calis et al., PLoS Comput Biol 2013, PMID 24204222). Requires `ligandome/viral_foreign_iedb.tsv.gz`; no cysteine artefact, but a 2x C-terminal Arg/Lys gradient |
+| `immunogenicity/kesmir_rebuilt_hla_matched.tsv.gz` | …HLA-matched | 59,941 rows, same matching rule |
+| `immunogenicity/iedb_labeled.tsv.gz` | IEDB immunogenicity — **ipred training set** | immunogenic (T-cell+, 34,362) vs non-immunogenic (MHC-ligand+ in healthy self, 756,506); `host_species` (human/mouse) is the cross-validation axis; ipred featurizes `peptide` (MJ/hydropathy/Kidera) |
+| `neoantigens/neoag_tested.tsv.gz` | Neoantigens — tested (workhorse) | tested patient-derived neoantigens, immunogenicity 0/1 (TESLA, Neopep, CEDAR, Gfeller, ITSNdb, GBM, VACCIMEL, Bjerregaard). **Human only by label — see the mouse-contamination caveat below.** |
+| `neoantigens/neoag_candidates.tsv.gz` | Neoantigens — candidates | untested neoantigen candidates (Neopep `not_tested`) |
+| `neoantigens/neoag_tested_mmu.tsv.gz` | Neoantigens — tested, **mouse MHC** | 1,977 murine tested neo-epitopes (IEDB), immunogenicity 0/1 (740 immunogenic); `mhc_species=MusMusculus` |
+| `neoantigens/neoag_tested_hsa.tsv.gz` | Neoantigens — tested, **human MHC in mouse models** | 717 rows from the same IEDB export whose restriction is HLA — HLA-transgenic mice plus human-host epitopes (KRAS G12D, CDK4 R24C, PMEL); `mhc_species=HomoSapiens` |
+| `neoantigens/nci_complete_minigenes.tsv.gz` | Neoantigens — complete NCI screen (variant) | 1,326 mutant 25-mer minigenes across 46 GI-cancer patients with CD8/CD4 T-cell screen outcomes (28 CD8-immunogenic) — a predictor-agnostic *screened* set with real negatives. Standalone variant, **minigene-level** (`peptide` = minigene, `mhc_a` blank — not epitope-deconvolved). Source: Koşaloğlu-Yalçın et al., iScience 2022, Table S6 ([doi:10.1016/j.isci.2022.103850](https://doi.org/10.1016/j.isci.2022.103850), PMID 35128348) |
+| `neoantigens/neoantigens_tested_peptides.tsv.gz` | Neoantigens — tested, **epitope-level** | 423,085 screened 8–12mers with their wild-type counterpart and restricting alleles, 178 CD8-immunogenic, 98 patients, over three datasets (`NCI` 420,786 · `HiTIDE` 1,563 · `TESLA` 736) and the authors' own `split`. This is the screen at the resolution a pMHC model is scored at — the complement to the minigene-level `nci_complete`. Harmonized from `raw/immunogenicity/NCI_dataset_only_tested.txt` by `bench/neoag/ingest_tested.py` (2026-mhcmatch-benchmark). Predicted quantities keep a `pred_` prefix; `peptide`, `peptide_wt`, `mhc_a` and `immunogenicity` are as-reported |
+| `neoantigens/nci_gartner_mmp.tsv.gz` | NCI — confirmed CD8+ minimal epitopes | 147 rows / 144 distinct minimal mutant peptides (mmp) with restricting HLA; 120 carry the wild-type mmp, so agretopicity is **measurable rather than predicted**. `split` preserves the paper's train/test. Source: Gartner et al., Nat Cancer 2021, Suppl. Tables 3 & 20 ([doi:10.1038/s43018-021-00197-6](https://doi.org/10.1038/s43018-021-00197-6)) |
+| `neoantigens/nci_gartner_nmers.tsv.gz` | NCI — full screened nmer universe | 9,541 screened mutations as WT/mutant 25-mers (144 CD8+), with gene, mutation type, RNA-seq presence and exome VAF decile — the negative universe at the unit the assay used. Same source, Suppl. Tables 11 & 12 |
+| `neoantigens/nci_gartner_samples.tsv.gz` | NCI — patient class-I genotypes | 98 patients with full HLA-A/B/C genotype, histology and per-patient screening counts. The genotype is what makes a per-patient allele-aware scan possible. Same source, Suppl. Tables 1 & 18 |
+| `tcga/tcga_neoantigens.tsv.gz` | TCGA neoantigens | 2.24M expressed mutant-peptide–HLA binders across 8,505 TCGA donors (derived from open-access TCGA somatic + expression; per-barcode HLA/neoantigens as in TCIA / Thorsson 2018) |
+| `immunogenicity/hla_pop_freqs.tsv.gz` | HLA population freqs | EUR/ASN/AFR class-I (A/B/C) allele frequencies (AFND) |
+| `proteome/human.fasta.gz` | Human proteome (UP000005640) | self-reference proteome for similarity search |
+| `proteome/mouse.fasta.gz` | Mouse proteome (UP000000589) | C57BL/6 self-reference proteome |
+| `proteome/*_UP*.fasta.gz` | 16 bacterial + viral reference proteomes | foreign-antigen references for molecular-mimicry scans and peptide-flank extraction; see `proteome/README.md` |
+| `thymus/thymus_immunopeptidome.tsv.gz` | Thymus self-peptidome (human) | HLA Ligand Atlas thymus-eluted self peptides (25.9k MHC-I + 28.0k MHC-II) — central-tolerance "self" reference for seqtree; fills the mislabeled source #1 |
+| `thymus/thymus_expression.tsv.gz` | Thymus expression (human) | thymically expressed self: HPA thymus-cluster genes (126) + atlas thymus source proteins (10.4k) |
+| `ligandome/viral_foreign_iedb.tsv.gz` | Viral ligandome (IEDB) | viral-source presented peptides (foreign reference) |
+| `ligandome/viral_orfs_gse272406.tsv.gz` | Pan-viral ORFs (GSE272406) | translated novel viral ORF proteins (foreign; *not* thymus self — spec mislabel) |
+| `ligandome/cancer_targets_tsarina.tsv.gz` | Cancer-testis antigens | curated shared tumor-antigen genes (tsarina) |
+| `summary.tsv` | Pipeline stats | row counts & extraction metrics |
 
 ## Private (gitignored — NOT in this repo)
 
 | Path | # | Dataset |
 |------|---|---------|
-| `gamaleya/patient_features.tsv.gz`, `gamaleya/hla_coverage.tsv.gz` | 8 | Gamaleya patient candidates + features (17 patients) — real cohort, never published |
+| `gamaleya/patient_features.tsv.gz`, `gamaleya/hla_coverage.tsv.gz` | Gamaleya patient candidates + features (17 patients) — real cohort, never published |
 
 An interactive, self-contained catalog of the public datasets (breakdowns by species, MHC
 class, source, category, tissue, and downstream tool) is in [`catalog.html`](catalog.html).

expression/SOURCES.md

@@ -0,0 +1,23 @@
+# `expression/` — sources
+
+One file, two references that are **never merged** — different measurements, different units, and a
+gene-keyed table and a peptide-keyed one are not the same object.
+
+| | |
+|---|---|
+| file | `reference_expression.tsv.gz` |
+| rows | 5,656,063 |
+| schema | `key`, `key_type`, `source`, `context`, `median_tpm`, `q25_tpm`, `q75_tpm`, `n` |
+| provenance | **experimental** (both sources are measured RNA-seq); the per-context median/IQR/`n` are **computed** summaries of it |
+
+| `source` | `key_type` | rows | contexts | what a row is |
+|---|---|--:|--:|---|
+| `gtex` | `gene` | 3,955,284 | 53 | a gene's expression across normal tissue — the **safety** read |
+| `tcga` | `peptide` | 1,700,779 | 19 | a mutant peptide's expression by tumour type — the **target** read |
+
+`key_type` is the join key and it differs by source on purpose: GTEx answers "is this gene on in
+normal tissue", TCGA answers "is this neoantigen transcribed in this tumour". A caller that treats
+them as one column is asking a question neither answers.
+
+Read through `mhcmatch.expression` (`lookup`, `tissues`, `tumor_types`, `safety_profile`), which
+keeps the two apart. Fetched by `mhcmatch bootstrap --reference` (~105 MB).

pmhc/SOURCES.md

@@ -0,0 +1,30 @@
+# `pmhc/` — sources
+
+The reference ligand panel every `mhcmatch.Store` is built from. **Experimental**: positive IEDB
+MHC-ligand records, harmonised; nothing here is predicted.
+
+| file | rows | what it is |
+|---|--:|---|
+| `pmhc_full.tsv.gz` | 1,482,188 | every positive IEDB epitope–MHC record |
+| `pmhc_shortlist.tsv.gz` | 645,102 | the subset supported by **≥2 references** |
+
+Schema: `epitope`, `gene`, `species`, `mhc_a`, `mhc_b`, `mhc_class`, `mhc_species`, `reference_id`,
+`n_references`.
+
+`pmhc_full.tsv.gz` composition:
+
+| class | host | rows |
+|---|---|--:|
+| MHCI | HomoSapiens | 1,040,631 |
+| MHCII | HomoSapiens | 327,487 |
+| MHCI | MusMusculus | 97,601 |
+| MHCII | MusMusculus | 16,469 |
+
+**The two tiers are a precision/recall choice, not a quality ranking.** `shortlist` drops
+single-reference pairs, which removes both the noise and the rare alleles that only one study ever
+reported — so it is the better background and the worse coverage. `full` is the default;
+`Store.from_pmhc(tier="shortlist")` selects the other.
+
+```zsh
+mhcmatch bootstrap --tier all      # fetches both, ~16 MB
+```

proteome/SOURCES.md

@@ -0,0 +1,26 @@
+# `proteome/` — sources
+
+**Provenance lives in [`README.md`](README.md)** beside this file, and is not repeated here: it
+carries every proteome's organism, UniProt proteome ID, taxid, sequence count and gzipped size, plus
+the exact fetch form (UniProt REST `?query=(proteome:<UPID>)&format=fasta`, or the
+reference_proteomes FTP). All 19 files are **experimental** sequence, unmodified.
+
+Summary: `human.fasta.gz` (UP000005640, 147,506 seqs) and `mouse.fasta.gz` (UP000000589, 54,857) are
+the **full** proteomes including isoforms and TrEMBL; the other 16 are canonical **reference**
+proteomes — 5 bacterial and 11 viral, 16,467 sequences in total — for molecular-mimicry scans and
+peptide-flank extraction.
+
+```zsh
+mhcmatch bootstrap --proteome human,mouse,ecoli_K12_UP000000625
+```
+
+Two traps the README records and that are easy to reintroduce:
+
+- **UP000007639 is EBV strain AG876, not HHV-7**, despite the adjacency of the IDs. The genuine
+  HHV-7 reference is UP000098510 (strain RK, 84 proteins).
+- Two EBV strains are present on purpose (B95-8 and AG876); they are not duplicates.
+
+**Size is the operational fact.** The human proteome has 68,389,335 nine-mer windows — every
+position of every protein, because a source lookup is a question about *where* — and one seqtree
+index over them costs ~12.6 GB. Ask for the lengths you need
+(`mhcmatch.proteome.Proteome.find_sources` builds one index per distinct query length).

tcga/SOURCES.md

@@ -0,0 +1,26 @@
+# `tcga/` — sources
+
+Pan-cancer neoantigen predictions over open-access TCGA. **Derived, not measured** — every row is a
+prediction, and the only experimental inputs behind it are the TCGA somatic calls, expression and
+per-barcode HLA genotypes.
+
+| file | rows | what it is |
+|---|--:|---|
+| `tcga_neoantigens.tsv.gz` | 2,244,779 | expressed mutant peptide × HLA binders, 8,505 donors |
+| `hla_coverage.tsv.gz` | 353 | per-allele donor counts, donor fraction, allele frequency, epitope counts |
+
+`tcga_neoantigens` schema: `sample`, `peptide`, `mhc_a`, `affinity`, `expression`, `ensp`,
+`mhc_class`, `mhc_species`, `source_species`, `host_species`, `immunogenicity`, `dataset_origin`,
+`cancer_type`.
+
+Derived from open-access TCGA somatic + expression data, with per-barcode HLA genotypes and
+neoantigen calls as published by TCIA / Thorsson et al. 2018. **No controlled-access data is used or
+reproduced here**, which is why this directory is public while the NCI intermediates (dbGaP
+phs001003) never leave local disk.
+
+**`immunogenicity` in this file is a prediction, not an assay result.** Nothing in TCGA was
+T-cell-screened. Use it as a candidate universe and a per-allele prevalence reference; a model
+validated against it has been validated against another model.
+
+`hla_coverage.tsv.gz` is what makes an allele-aware scan honest: it says how many donors carry each
+allele, so a per-allele result can be weighted by how much of the cohort it actually covers.

thymus/SOURCES.md

@@ -0,0 +1,43 @@
+# `thymus/` — sources
+
+The central-tolerance reference: what a developing T-cell repertoire was negatively selected
+against. Both files are **experimental** (mass spectrometry / antibody-based expression); the
+harmonisation into this schema is computed.
+
+## `thymus_immunopeptidome.tsv.gz` — 53,878 rows
+
+Thymus-eluted self peptides, **HLA Ligand Atlas** (`dataset_origin = hla_ligand_atlas_thymus`).
+
+| `mhc_class` | rows |
+|---|--:|
+| MHCII | 27,987 |
+| MHCI | 25,891 |
+
+Schema: `peptide`, `mhc_a`, `mhc_class`, `mhc_species`, `source_protein`, `species`,
+`dataset_origin`. Human only.
+
+**A hit here is a tolerance argument and an autoimmunity flag, and those are the same fact read two
+ways**: a neoantigen resembling a thymically presented peptide had its reactive clones plausibly
+deleted (lower expected immunogenicity), and a vaccine built on it risks cross-reacting with self.
+`mhcmatch.mimics` keeps it as its own category for exactly that reason and never sums it with the
+foreign ones — see `mhcmatch.mimics.KINDS`.
+
+## `thymus_expression.tsv.gz` — 10,493 rows
+
+Thymically expressed self, by gene. Schema: `gene`, `species`, `evidence`, `value`.
+
+| `evidence` | rows | what it is |
+|---|--:|---|
+| `hla_ligand_atlas_thymus_source` | 10,367 | source proteins of the peptides above |
+| `hpa_thymus_cluster` | 126 | Human Protein Atlas thymus-cluster genes |
+
+The two lines of evidence are kept in a column rather than pooled: one is "a peptide from this
+protein was eluted from thymus", the other is "this gene is thymus-enriched by expression". They
+support each other; they are not interchangeable.
+
+## Not here
+
+**No mouse thymus reference.** The candidate deposits do not survive contact: the mouse thymus
+proteome (PXD007288) is a 12 GB MaxQuant archive and the mTEC-TRA gene list (Sansom, GSE53111) is
+journal-supplement-only, so a mouse pass means reprocessing raw spectra. Recorded in
+`DESCRIPTION.md` and open in the benchmark repo's TODO — **absent, not overlooked**.