Extend catalog to 238 predictors for pyaging 0.5.5

README.md

@@ -15,9 +15,13 @@ This public repository contains the model weights and data files used by
 
 ## Contents
 
-- Root-level `*.pt` files are the current pyaging clock models.
-- `all_clock_metadata.pt` is the live aggregate clock catalog.
-- Clock-specific `*.provenance.json` and `*.LICENSE.txt` files record sources and supplied author terms.
+- Clock model weights live in one repository per clock under the
+  [`pyaging` organization](https://huggingface.co/pyaging) (e.g.
+  `pyaging/horvath2013`); each repo carries the weight file, the audited clock
+  metadata as `config.json`, and a model card. Root-level `*.pt` files here are
+  compatibility copies used when a per-clock repository or revision is unavailable.
+- `all_clock_metadata.pt` is the live aggregate clock catalog. Retired clock files
+  may remain here to preserve earlier releases; file presence is not current catalogue membership.
 - Root-level example files support the pyaging tutorials.
 - `supporting_files/` contains dependencies used to construct or document clocks.
 
@@ -26,6 +30,21 @@ downloaded through the standard Hugging Face cache.
 The `main` branch is the live data release and may change independently of the Python
 package version.
 
+## Versioning and reproducibility
+
+Each pyaging release tags this repository with the matching package version (e.g.
+`v0.3.1`), so a release tag captures the exact data files that shipped with that
+version of the package. By default pyaging downloads from `main`; set the
+`PYAGING_DATA_REVISION` environment variable to a release tag (or any commit) to pin
+downloads to that revision:
+
+```bash
+PYAGING_DATA_REVISION=v0.3.1 python my_analysis.py
+```
+
+Release tags only exist for pyaging versions published after the tagging scheme was
+introduced.
+
 ## Harmonized clock metadata
 
 The clock catalogue uses controlled, multi-valued metadata so clocks can be
@@ -51,7 +70,7 @@ are maintained in the pyaging repository.
 ## Licensing and provenance
 
 This is a mixed-provenance research collection, so the repository license is `other`.
-The pyaging software MIT license does not grant additional rights to third-party clock weights or
+The pyaging MIT license does not grant additional rights to third-party clock weights or
 source datasets. Consult each clock's embedded metadata, cited publication, and notes
 before use. Some clocks are marked research-only or have separate commercial terms.
 
@@ -66,3 +85,17 @@ load these files from this official repository and review unexpected repository
 The repository is maintained solely by Lucas Paulo de Lima Camillo (`lucascamillomd`).
 Weights are uploaded before aggregate metadata so the catalog never advertises a missing
 clock file. Public users need no Hugging Face token to download files.
+
+## Proteomic inputs
+
+The 0.5.5 catalogue contains PAC and 46 full Olink Explore 3072 OrganAge models.
+Their input units and protein identifiers are model-specific. OrganAge uses
+case-sensitive original symbols and PAC uses lowercase symbols plus age in years.
+NPX is log2 relative abundance, not concentration; cross-platform and serum/plasma
+harmonization are external preparation decisions. See the
+[proteomic input guide](https://pyaging.readthedocs.io/en/latest/proteomic_clocks.html).
+
+The reduced Olink Explore 1536 OrganAge models leave the current catalogue, and
+full-model names drop `olink3000`. Existing 0.5.4 tags and files remain intact.
+ProtAge, ipfP3GPT and PAOPAC are not executable pyaging entries in this release;
+the guide explains the unavailable or restricted author assets.

all_clock_metadata.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7295b4383b726abfc0db9a44d47ef1209452a70044340424906b74891886dc7d
-size 329775
+oid sha256:72a2da13c489777f089efa61048ec6c2fc7185db731836cb208ca83e9e33c247
+size 279279