neoantigens: the rest of the Sahin TNBC supplementary, processed

CC BY 4.0 (verified against Crossref), so the processed tables are redistributable with
attribution; the source xlsx stay out. Adds the driver-gene responses, the 24,517-gene
DESeq2 tumour-vs-normal contrast and its GSEA. The DEG table is log2 fold change, not TPM,
and is named so that stays visible -- it is not a drop-in for the expr feature.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

neoantigens/SOURCES.md

@@ -168,3 +168,32 @@ two, not four, in any leave-one-cohort-out design.
 
 `neoag_tested*` / `neoag_candidates` / `nci_complete_minigenes` are deposited as fetched and are not
 regenerated by this repository.
+
+## Sahin TNBC 2025 — the full supplementary set (2026-08-19)
+
+Sahin et al., *Nature* 2025, "Individualized mRNA vaccines evoke durable T cell immunity in adjuvant
+TNBC", [doi:10.1038/s41586-025-10004-2](https://doi.org/10.1038/s41586-025-10004-2). Published under
+**CC BY 4.0**, verified against Crossref, so the *processed* tables are redistributable here with
+attribution. The source `.xlsx` and the article PDF stay in `~/hf/tnbc` and are **not** deposited.
+
+| file | from | rows | provenance |
+|---|---|--:|---|
+| `sahin_tnbc_2025.tsv.gz` | MOESM3, sheet `n=251 targets` | 251 | **experimental** — one row per vaccinated target; `ex_vivo` and `post_ivs` are two separate ELISpot read-outs, `mean_spot_count` the magnitude |
+| `sahin_tnbc_2025_drivers.tsv.gz` | MOESM3, sheet `Breast cancer driver genes` | 6 | **experimental** — driver-gene targets and their per-patient responses |
+| `sahin_tnbc_2025_deg.tsv.gz` | MOESM4 | 24,517 | **computed** — DESeq2 tumour vs normal on the same cohort: `baseMean, log2FoldChange, lfcSE, stat, pvalue, padj` |
+| `sahin_tnbc_2025_gsea.tsv.gz` | MOESM5 | 43 | **computed** — gene-set enrichment over those genes |
+
+`sahin_tnbc_2025_deg.tsv.gz` is a **log2 fold change, not TPM**, and is deposited under its own name
+so that distinction stays visible: it is not a drop-in for the `expr` feature, which is an absolute
+expression level. It is the only tumour-vs-normal contrast in this directory measured on the same
+patients as a prospective holdout.
+
+This cohort is **held out by construction** — it was published after every model in the benchmark
+repository was fitted, and enters no fit.
+
+Regenerate:
+
+```zsh
+python bench/neoag/ingest_sahin.py         # -> sahin_tnbc_2025.tsv.gz
+python bench/neoag/ingest_sahin_extra.py   # -> the drivers, deg and gsea tables
+```

neoantigens/sahin_tnbc_2025_deg.tsv.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bec570eef94b1f1e3bf1a06b15dba9643ea33182865b35ec1d6efbf8b0771de
+size 1392118

neoantigens/sahin_tnbc_2025_drivers.tsv.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea2ff9cf00dcc8f6d21c7a7662155c18d9bc3b4724f6018376c7ccbff47e261c
+size 301

neoantigens/sahin_tnbc_2025_gsea.tsv.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e93e8d6bbf33736aa11cf24cfe90e9a7a86dae0775ea1c2f3c4e5114a5339ec9
+size 6975