docs: add immutable resource card
Browse files
README.md
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: NPPI-Net reproducibility resources
|
| 3 |
+
license: other
|
| 4 |
+
task_categories:
|
| 5 |
+
- tabular-regression
|
| 6 |
+
tags:
|
| 7 |
+
- natural-products
|
| 8 |
+
- protein-ligand-interaction
|
| 9 |
+
- molecular-embeddings
|
| 10 |
+
- protein-embeddings
|
| 11 |
+
size_categories:
|
| 12 |
+
- 10K<n<100K
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# NPPI-Net reproducibility resources
|
| 16 |
+
|
| 17 |
+
This dataset repository contains the exact large inputs used by the retained
|
| 18 |
+
NPPI-Net S1-S4 result chain. Source code, fixed split manifests, paper
|
| 19 |
+
configurations, raw metrics, and verification scripts are published separately
|
| 20 |
+
in
|
| 21 |
+
[`Cai-z-us/natural-product-protein-interact`](https://github.com/Cai-z-us/natural-product-protein-interact).
|
| 22 |
+
Paper tables and images are not duplicated here.
|
| 23 |
+
|
| 24 |
+
## Files
|
| 25 |
+
|
| 26 |
+
| Path | Purpose | Bytes | SHA256 |
|
| 27 |
+
| --- | --- | ---: | --- |
|
| 28 |
+
| `data/dataset_all.csv` | canonical headerless interaction dataset | 23,281,585 | `0bef27914ed06e5fe0af9ad64ecd16f27d007369b738f65240795505708d8017` |
|
| 29 |
+
| `data/external/uniprot_human_reviewed.fasta` | exact historical reviewed-human UniProtKB snapshot | 13,663,585 | `89b398077ec571c34bd67fd10fc1640eeae37d1421d6f84c3a8a94e1db88dac3` |
|
| 30 |
+
| `data/mol/unimol_v2_1.1B.pth` | pooled Uni-Mol2 molecule representations | 1,460,616,917 | `f87c3673e911e1cc21022e4c4ddcb40cfb2b9ee4e355899ba02f60ae21ad83da` |
|
| 31 |
+
| `data/prot/proteinglm_xtmlm_1b.pth` | pooled ProteinGLM protein representations | 9,691,794,119 | `1abb24b598db851733b7345618284aab78ef0ce4d431b454878fa63451d20cb4` |
|
| 32 |
+
|
| 33 |
+
The CSV has 32,802 rows, 8,959 unique molecule identifiers, 1,860 unique
|
| 34 |
+
protein identifiers, and no header. Its ten columns are, in order:
|
| 35 |
+
`mol_id`, `smiles`, `uniprot_id`, `sequence`, `pchembl`,
|
| 36 |
+
`standard_relation`, `standard_value_nM`, `standard_units`, `standard_type`,
|
| 37 |
+
and `confidence_score`.
|
| 38 |
+
|
| 39 |
+
The retained model consumes pooled `float32` vectors only: 1,536 dimensions
|
| 40 |
+
for molecules and 2,048 for proteins. Token inputs are disabled in the paper
|
| 41 |
+
configuration.
|
| 42 |
+
|
| 43 |
+
## Download and verify
|
| 44 |
+
|
| 45 |
+
From the root of a clone of the source repository:
|
| 46 |
+
|
| 47 |
+
```bash
|
| 48 |
+
hf download Caizus/natural-product-protein-interact \
|
| 49 |
+
--repo-type dataset \
|
| 50 |
+
--revision d67ea7b4f75b35f1f362caebcd3d5565cedc16dc \
|
| 51 |
+
--local-dir .
|
| 52 |
+
sha256sum --check SHA256SUMS
|
| 53 |
+
uv run --frozen python scripts/verify_resources.py --deep
|
| 54 |
+
```
|
| 55 |
+
|
| 56 |
+
`scripts/verify_resources.py` checks byte size, SHA256, schema, key alignment,
|
| 57 |
+
pooled dimensions, and dtypes against
|
| 58 |
+
`artifacts/resources/manifest.yaml`. PyTorch files are opened with the
|
| 59 |
+
restricted `weights_only` loader.
|
| 60 |
+
|
| 61 |
+
## Provenance
|
| 62 |
+
|
| 63 |
+
- The canonical interaction dataset was built from ChEMBL 34 and a reviewed
|
| 64 |
+
human UniProtKB snapshot. The exact FASTA is published above because the
|
| 65 |
+
historical UniProt release response header was not retained; its byte
|
| 66 |
+
identity, rather than an inferred release number, is authoritative.
|
| 67 |
+
- The molecule file was generated with `unimol-tools==0.1.5` and Uni-Mol2
|
| 68 |
+
1.1B. The package wheel and final output are content-addressed in the source
|
| 69 |
+
repository.
|
| 70 |
+
- The protein file was generated from
|
| 71 |
+
`biomap-research/proteinglm-1b-mlm` at revision
|
| 72 |
+
`676c1bcb9f737c166a25587f83fd580610acb1ad`.
|
| 73 |
+
|
| 74 |
+
See
|
| 75 |
+
[`ARTIFACTS.md`](https://github.com/Cai-z-us/natural-product-protein-interact/blob/main/ARTIFACTS.md),
|
| 76 |
+
[`DATASET.md`](https://github.com/Cai-z-us/natural-product-protein-interact/blob/main/DATASET.md),
|
| 77 |
+
and
|
| 78 |
+
[`THIRD_PARTY_NOTICES.md`](https://github.com/Cai-z-us/natural-product-protein-interact/blob/main/THIRD_PARTY_NOTICES.md)
|
| 79 |
+
for the full build and attribution record.
|
| 80 |
+
|
| 81 |
+
## Terms
|
| 82 |
+
|
| 83 |
+
This repository contains artifacts with different governing terms; the
|
| 84 |
+
project's future source-code license does not replace them.
|
| 85 |
+
|
| 86 |
+
- ChEMBL-derived dataset content is subject to ChEMBL 34's CC BY-SA 3.0 terms
|
| 87 |
+
and required attribution.
|
| 88 |
+
- UniProt-derived content is subject to CC BY 4.0 attribution.
|
| 89 |
+
- Uni-Mol tooling/model attribution is retained; applicable source-data terms
|
| 90 |
+
continue to apply to the generated molecule representations.
|
| 91 |
+
- ProteinGLM is CC BY-NC 4.0. The generated protein representations must be
|
| 92 |
+
treated as noncommercial material and must not be advertised as unrestricted
|
| 93 |
+
commercial assets.
|
| 94 |
+
|
| 95 |
+
No additional artifact-wide license is asserted beyond those file-specific
|
| 96 |
+
terms while the authors finalize the paper release metadata.
|