Initialize verified RAW 70%-cluster release metadata

BUILD_RECIPE.md

@@ -0,0 +1,117 @@
+# Reproducing the 70%-identity representative FASTAs
+
+The canonical implementation is
+[`dev/data/process_full_corpus.py`](https://github.com/Lumin-Science/LuminBench-Nano-ESMC/blob/main/dev/data/process_full_corpus.py).
+All Python environments are managed by `uv`; MMseqs2 is invoked as a pinned
+external binary and its version is recorded in each clustering receipt.
+
+## 1. Pin and download the three source arms
+
+| Source | Frozen input |
+|---|---|
+| UniRef90 | [UniRef release 2023_02 archive](https://ftp.uniprot.org/pub/databases/uniprot/previous_releases/release-2023_02/uniref/uniref2023_02.tar.gz), 211,819,312,677 bytes, MD5 `353681f464572bb199fa032f714d4669` |
+| MGnify | [Protein DB 2023_02 `mgy_clusters.fa.gz`](https://ftp.ebi.ac.uk/pub/databases/metagenomics/peptide_database/2023_02/mgy_clusters.fa.gz), 83,473,342,442 bytes, MD5 `332d36d2a943bdb769237a03e050ed03` |
+| OMG/IMG | All 959 Parquet objects in [`tattabio/OMG`](https://huggingface.co/datasets/tattabio/OMG), 1,253,813,127,320 bytes total, pinned by path, size, and LFS SHA-256 in [`dev/data/omg_upstream_shards.tsv`](https://github.com/Lumin-Science/LuminBench-Nano-ESMC/blob/main/dev/data/omg_upstream_shards.tsv) |
+
+## 2. Normalize and quality-filter
+
+For every record, in order:
+
+1. remove whitespace and a terminal `*`;
+2. uppercase the sequence;
+3. map unsupported amino-acid characters to `X`;
+4. reject sequences shorter than 60 residues;
+5. reject sequences with more than 20% non-canonical residues; and
+6. in the OMG payload, keep numeric JGI/IMG accessions and remove `ERZ` MGnify-
+   origin rows so MGnify is not represented twice.
+
+The measured funnel was:
+
+| Source | Original records | Short | >20% ambiguous | Wrong source | Quality eligible |
+|---|---:|---:|---:|---:|---:|
+| UniRef90 | 170,669,877 | 4,755,787 | 29,797 | 0 | 165,884,293 |
+| MGnify | 729,215,663 | 117,372,608 | 26,021 | 0 | 611,817,034 |
+| OMG/IMG | 3,280,269,924 | 145,672,826 | 0 | 1,048,850,563 | 2,085,746,535 |
+
+## 3. Global exact deduplication
+
+Accepted records are partitioned into 256 leading-byte SHA-256 buckets. Exact
+normalized sequences are collapsed globally while preserving every
+`(sha256, source, source_id)` membership in Parquet. A source-specific FASTA
+view is then emitted for every source in which the sequence occurred.
+
+The source views contain 165,884,293 UniRef90, 611,788,129 MGnify, and
+963,673,186 OMG/IMG unique sequences. Their global union contains
+1,661,993,387 exact unique sequences; source-view counts are larger because one
+digest can belong to multiple source arms.
+
+## 4. Source-specific diversity clustering
+
+Each source view is clustered independently with MMseqs2 Linclust:
+
+```text
+mmseqs linclust sequences clusters tmp \
+  --min-seq-id 0.70 \
+  -c 0.80 \
+  --cov-mode 1 \
+  --cluster-mode 2 \
+  --threads 64
+```
+
+`--cov-mode 1` applies the 80% coverage requirement to the shorter sequence.
+`createtsv` emits the representative/member digest pairs and `result2repseq`
+plus `result2flat --use-fasta-header` emits the representative FASTA.
+
+## 5. Reproduction commands
+
+```bash
+uv sync --frozen
+PIPE=dev/data/process_full_corpus.py
+ROOT=/absolute/path/to/protein-corpus
+MMSEQS=/absolute/path/to/mmseqs
+
+uv run --frozen python "$PIPE" download \
+  --data-root "$ROOT" \
+  --omg-manifest dev/data/omg_upstream_shards.tsv \
+  --download-workers 8
+
+uv run --frozen python "$PIPE" normalize --source uniref90 \
+  --input "$ROOT/raw/uniref90_2023_02/uniref2023_02.tar.gz" \
+  --output "$ROOT/normalized/uniref90"
+
+uv run --frozen python "$PIPE" normalize --source mgnify \
+  --input "$ROOT/raw/mgnify_2023_02/mgy_clusters.fa.gz" \
+  --output "$ROOT/normalized/mgnify"
+
+# Expand all 959 OMG inputs from dev/data/omg_upstream_shards.tsv, then:
+uv run --frozen python "$PIPE" normalize --source omg_img \
+  "${omg_inputs[@]}" --output "$ROOT/normalized/omg_img"
+
+uv run --frozen python "$PIPE" deduplicate \
+  --input "$ROOT/normalized/uniref90" \
+  --input "$ROOT/normalized/mgnify" \
+  --input "$ROOT/normalized/omg_img" \
+  --output "$ROOT/deduplicated"
+
+for source in uniref90 mgnify omg_img; do
+  uv run --frozen python "$PIPE" cluster \
+    --dedup-root "$ROOT/deduplicated" \
+    --source "$source" \
+    --output "$ROOT/clusters/$source" \
+    --mmseqs "$MMSEQS" \
+    --threads 64
+done
+```
+
+Every output directory is create-once. The builder refuses to overwrite an
+existing normalization, deduplication, or clustering directory.
+
+## 6. Boundary of this RAW release
+
+This release stops immediately after source-specific 70%-identity clustering.
+It does not perform evaluation homology exclusion, exact evaluation exclusion,
+cross-source representative ownership, length 32–16,384 filtering, validation
+selection, or Parquet packing. Those steps produce the final
+[`LuminBench-Nano-ESMC`](https://huggingface.co/datasets/LuminScience/LuminBench-Nano-ESMC)
+release.
+

LICENSE_AND_ATTRIBUTION.md

@@ -0,0 +1,48 @@
+# License and attribution
+
+Reviewed: 2026-08-26
+
+This repository contains a mixed-terms compilation. Lumin Science licenses only
+its original selection, arrangement, SHA-addressed clustering metadata,
+verification receipts, and documentation under
+[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/). It does not
+relicense third-party protein sequences or identifiers.
+
+## UniRef90
+
+`representatives/uniref90.fasta` derives from UniRef release 2023_02. UniProt
+applies [CC BY 4.0](https://www.uniprot.org/help/license) to copyrightable
+database content. Attribute the UniProt Consortium / UniRef, retain its notice,
+link to the source where practical, and state the processing changes.
+
+## MGnify
+
+`representatives/mgnify.fasta` derives from the official MGnify Protein Database
+2023_02 archive. It remains governed by the
+[EMBL-EBI Terms of Use](https://www.ebi.ac.uk/about/terms-of-use/) and applicable
+original-owner rights. Attribute EMBL-EBI MGnify and the original contributors.
+Lumin Science does not apply a new Creative Commons license to these records.
+
+## OMG/IMG
+
+`representatives/omg_img.fasta` contains numeric-accession JGI/IMG protein CDS
+rows from [`tattabio/OMG`](https://huggingface.co/datasets/tattabio/OMG), after
+removing its `ERZ` MGnify-origin rows. The direct distribution declares CC
+BY-SA 4.0. Attribute the OMG authors/TattaBio, JGI/IMG, and original
+contributors; retain the license and modification notices.
+
+## Derived cluster maps
+
+`clusters/**`, receipts, manifests, and build documentation are Lumin Science
+derived database metadata offered under CC BY-SA 4.0. The cluster maps contain
+SHA-256 digests rather than member sequences or original accessions. Applying
+the maps to upstream data does not replace the upstream terms for the recovered
+records.
+
+## Processing changes
+
+Records were normalized, quality filtered, globally exact-deduplicated with
+source membership retained, and clustered independently per source at 70%
+sequence identity and 80% shorter-sequence coverage. The data in this RAW
+repository have not yet been evaluation-decontaminated.
+

README.md

@@ -0,0 +1,91 @@
+---
+pretty_name: LuminBench Nano ESMC Raw 70% Cluster Outputs v1
+license: other
+size_categories:
+  - 100M<n<1B
+tags:
+  - biology
+  - protein
+  - protein-language-model
+  - fasta
+  - clustering
+  - mmseqs2
+---
+
+# LuminBench Nano ESMC Raw 70% Cluster Outputs v1
+
+This repository preserves the source-specific 70%-identity clustering outputs
+that precede evaluation decontamination and final Parquet packing in
+[`LuminScience/LuminBench-Nano-ESMC`](https://huggingface.co/datasets/LuminScience/LuminBench-Nano-ESMC).
+
+It contains **241,600,826,413 bytes of representative FASTA data** covering
+**765,290,002 source-specific cluster representatives**. It also preserves the
+three cluster-membership maps so the clustering result is not reduced to the
+representatives alone.
+
+> **Not decontaminated training data.** These FASTAs precede the P@L/P-CORE
+> evaluation exclusion, cross-source ownership assignment, length gate, and
+> train/validation split. Use the final repository above for clean training.
+
+## Files
+
+| Source arm | Representative FASTA | Representatives | Cluster map | Source-specific member rows |
+|---|---:|---:|---:|---:|
+| UniRef90 2023_02 | 38,626,189,542 bytes | 92,230,941 | 23,887,338,192 bytes | 165,884,293 |
+| MGnify Protein DB 2023_02 | 93,319,863,106 bytes | 348,135,082 | 88,097,490,576 bytes | 611,788,129 |
+| OMG/IMG | 109,654,773,765 bytes | 324,923,979 | 138,768,938,784 bytes | 963,673,186 |
+| **Total** | **241,600,826,413 bytes** | **765,290,002** | **250,753,767,552 bytes** | **1,741,345,608 source memberships** |
+
+Paths are:
+
+```text
+representatives/{uniref90,mgnify,omg_img}.fasta
+clusters/{uniref90,mgnify,omg_img}.clusters.tsv
+receipts/{uniref90,mgnify,omg_img}.verification.json
+```
+
+FASTA identifiers are `sha256_<digest>`, where the digest is computed from the
+normalized ASCII amino-acid sequence. Each cluster-map row is:
+
+```text
+representative_sha256<TAB>member_sha256
+```
+
+The maps are source-specific. The same exact sequence can therefore occur as a
+member in more than one source arm.
+
+## Does this preserve every clustered protein?
+
+It preserves the **cluster assignment relation**, but it is not a self-contained
+copy of every cluster-member sequence:
+
+- the representative FASTAs contain the 765.29M representative sequences;
+- the `clusters.tsv` files map each representative digest to every member digest;
+- non-representative member sequences and original upstream accessions are not
+  stored in those TSVs.
+
+To recover all member sequences or original identifiers, redownload the pinned
+upstream snapshots, rerun normalization and exact deduplication, and join the
+resulting membership Parquet rows to `member_sha256`. This repository therefore
+preserves enough information to reproduce cluster membership **when combined
+with the pinned upstream inputs and build recipe**, but the representative
+FASTAs alone would not be sufficient.
+
+See [`BUILD_RECIPE.md`](BUILD_RECIPE.md), [`manifest.json`](manifest.json), and
+[`SOURCE_PROVENANCE.json`](SOURCE_PROVENANCE.json) for the exact pipeline,
+commands, checksums, and limitations.
+
+## License and source terms
+
+This is a mixed-terms dataset, so the Hub metadata uses `license: other` rather
+than pretending that one license replaces all upstream terms.
+
+| Path | Direct source | Governing terms |
+|---|---|---|
+| `representatives/uniref90.fasta` | UniRef90 2023_02 | [UniProt CC BY 4.0](https://www.uniprot.org/help/license) |
+| `representatives/mgnify.fasta` | MGnify Protein DB 2023_02 | [EMBL-EBI Terms of Use](https://www.ebi.ac.uk/about/terms-of-use/) plus applicable original-owner rights; not relicensed by Lumin Science |
+| `representatives/omg_img.fasta` | JGI/IMG records distributed by [`tattabio/OMG`](https://huggingface.co/datasets/tattabio/OMG) | CC BY-SA 4.0 as declared by the direct distribution |
+| `clusters/**`, manifests, receipts, and build documentation | Lumin Science selection, arrangement, and derived clustering metadata | CC BY-SA 4.0 |
+
+See [`LICENSE_AND_ATTRIBUTION.md`](LICENSE_AND_ATTRIBUTION.md) before reuse.
+

SOURCE_PROVENANCE.json

@@ -0,0 +1,32 @@
+{
+  "distribution_license": {
+    "license": "mixed; see LICENSE_AND_ATTRIBUTION.md",
+    "lumin_science_metadata_license": "CC BY-SA 4.0",
+    "third_party_records_relicensed": false
+  },
+  "release": "raw-source-specific-70pct-clusters-v1",
+  "reviewed_at": "2026-08-26",
+  "schema_version": 1,
+  "source_arms": {
+    "mgnify": {
+      "governing_terms": "EMBL-EBI Terms of Use plus applicable original-owner rights",
+      "raw_bytes": 83473342442,
+      "raw_md5": "332d36d2a943bdb769237a03e050ed03",
+      "raw_url": "https://ftp.ebi.ac.uk/pub/databases/metagenomics/peptide_database/2023_02/mgy_clusters.fa.gz"
+    },
+    "omg_img": {
+      "governing_terms": "CC BY-SA 4.0 as declared by tattabio/OMG",
+      "raw_bytes": 1253813127320,
+      "raw_manifest_sha256": "bb24ae4e819c92faa20767bf9f4070dfd3e27f673e77d064b7a45cc897ebd7d0",
+      "raw_repository": "https://huggingface.co/datasets/tattabio/OMG",
+      "raw_shards": 959
+    },
+    "uniref90": {
+      "governing_terms": "CC BY 4.0",
+      "raw_bytes": 211819312677,
+      "raw_md5": "353681f464572bb199fa032f714d4669",
+      "raw_url": "https://ftp.uniprot.org/pub/databases/uniprot/previous_releases/release-2023_02/uniref/uniref2023_02.tar.gz"
+    }
+  }
+}
+

manifest.json

@@ -0,0 +1,71 @@
+{
+  "clustering": {
+    "cluster_mode": 2,
+    "coverage_mode": 1,
+    "minimum_coverage": 0.8,
+    "minimum_sequence_identity": 0.7,
+    "method": "MMseqs2 Linclust",
+    "scope": "independent per source arm"
+  },
+  "files": {
+    "mgnify": {
+      "cluster_map": {
+        "bytes": 88097490576,
+        "path": "clusters/mgnify.clusters.tsv",
+        "rows": 611788129,
+        "sha256": "7c1ffedd2d772b6955da746f53ede3885e23242faad27fcae75a7323877d5342"
+      },
+      "representatives": {
+        "bytes": 93319863106,
+        "path": "representatives/mgnify.fasta",
+        "records": 348135082,
+        "sha256": "c40a97fcc8fee4777e8da8568e8f049f4e3f80dcecdc51f15da51005b28ad96a"
+      }
+    },
+    "omg_img": {
+      "cluster_map": {
+        "bytes": 138768938784,
+        "path": "clusters/omg_img.clusters.tsv",
+        "rows": 963673186,
+        "sha256": "530e00984d0f3eff9ca53a8588221b08a2bc0eb9ab5ce9b5302366289bbd4bbf"
+      },
+      "representatives": {
+        "bytes": 109654773765,
+        "path": "representatives/omg_img.fasta",
+        "records": 324923979,
+        "sha256": "163b764c9aca5241da1c499e58dcf25fa2b6670046bf1f2dbd28789aaa3e439e"
+      }
+    },
+    "uniref90": {
+      "cluster_map": {
+        "bytes": 23887338192,
+        "path": "clusters/uniref90.clusters.tsv",
+        "rows": 165884293,
+        "sha256": "50bede91d10a00b3c2c11ad22e16b87bf69cdecb96470e8a485382c869f16dd5"
+      },
+      "representatives": {
+        "bytes": 38626189542,
+        "path": "representatives/uniref90.fasta",
+        "records": 92230941,
+        "sha256": "521e57d099102945e11db852cc4f4870e288c2994c6a0a5dea9bff8e5bfdcec8"
+      }
+    }
+  },
+  "limitations": {
+    "cluster_maps_contain_member_sequences": false,
+    "cluster_maps_contain_original_accessions": false,
+    "evaluation_decontaminated": false,
+    "full_members_can_be_reconstructed_from_pinned_upstream_and_recipe": true
+  },
+  "protocol": "raw-source-specific-70pct-cluster-release-v1",
+  "schema_version": 1,
+  "status": "verified",
+  "totals": {
+    "cluster_map_bytes": 250753767552,
+    "representative_bytes": 241600826413,
+    "representative_records": 765290002,
+    "source_specific_member_rows": 1741345608,
+    "total_published_bytes": 492354593965
+  }
+}
+