neoag: primary datasets take precedence over CEDAR/Neopep; trim blank cols from iedb_labeled

- harmonize_neoag: drop CEDAR/Neopep rows whose peptide-MHC is already in a primary
dataset (273 dropped) -> neoag_tested 525,989, neoag_candidates 1,355,926
- iedb_labeled: drop always-empty feature columns (expression/affinity/affinity_wt/
agretopicity/foreignness/validated); 12 columns, 790,868 rows unchanged

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

DESCRIPTION.md

@@ -10,7 +10,10 @@ All tables are gzip-TSV. The immunogenicity tables share one schema: `peptide, m
 mhc_class, mhc_species, source_species, host_species, assay_type, immunogenicity{1,0,null},
 expression, affinity, affinity_wt, agretopicity, foreignness, validated, dataset_origin,
 reference_id, n_references` (+ `patient_id, cancer_type` for the neoag tables). `affinity` is
-as-reported per source (units differ); MHC alleles are as-reported.
+as-reported per source (units differ); MHC alleles are as-reported. `iedb_labeled.tsv.gz`
+omits the feature columns IEDB doesn't populate. In `neoag_tested`/`neoag_candidates`,
+primary datasets take precedence over the aggregated compendia (CEDAR, Neopep): a duplicate
+peptide–MHC already present in a primary dataset is dropped from CEDAR/Neopep.
 
 ## Public datasets (this repo)
 

immunogenicity/iedb_labeled.tsv.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f91e95566d4c6fabf02cdcdc8eb3f17ce5eb6afe7e51944eb24e41df0c08459f
-size 9893733
+oid sha256:eccc87b7108a0131c5b39f65629dc16950820599b5208503f9d40f356b8183b9
+size 9833065

immunogenicity/neoag_candidates.tsv.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d0b0e00c10c56c27a79722507e0e21ae35a8bce8c5181034cabadc9ce8a547b
-size 17532582
+oid sha256:54851cf9d574e0fe085a2f80c3f3c06cd84c60805612bb1e08da48e29d671d3c
+size 17531721

immunogenicity/neoag_tested.tsv.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:118658608f7046d5f007a7fb76ef3716eb4467a73474b8b32ffb7e87f44d6167
-size 7119906
+oid sha256:422ba89167bf327e8251f9bf8bb2c928435ad77b9706a724c5bab83e32e6540d
+size 7053481