Opal EmeraldBay
Observed-single expression reference for registered EmeraldBay chemical contexts
Quick start 路 Evaluation 路 Limitations 路 Provenance
Research preprint 路 Source code 路 The Tiny Intelligence Lab
Opal EmeraldBay is a historical contextual expression reference for chemical combinations with measured singles and registered cellular contexts. It combines source-local reference effects with a fitted contextual kernel and returns a population-mean expression-effect vector in the stored gene order.
Model and inputs
Use exact stored chemical/dose/time identifiers and a registered context. The example uses Adagrasib@0.005uM:5d and Capecitabine@0.5uM:5d in CVCL_0320. These are chemical interventions, not gene knockouts. An unavailable single, unknown intervention, unsupported context or mismatched gene order is rejected. The returned scale comes from the stored reference metadata.
Quick start
Use Python 3.11-3.14; the local verification used Python 3.14.0. Install CPU Torch first if required, followed by the pinned source and dependencies:
python -m pip install "torch==2.9.1" --index-url https://download.pytorch.org/whl/cpu
python -m pip install -r requirements.txt
python verify_files.py
python example.py
The requirements pin the exact Opal code revision. The source repository and v1.0.0 release are public. No model training runs during installation or prediction. These checkpoints use Opal's custom loaders, not transformers.AutoModel.
example.py checks the original manifest hash, loads the files in this repository, runs a supported pair query and verifies symmetry and finite output. A successful example is an inference smoke check, not a new biological evaluation.
Evaluation
This repository distributes a historical source-specific reference checkpoint. Packaging verifies original file hashes, successful inference, finite outputs, output gene order and pair-order symmetry. It performs no new fitting, hyperparameter selection or biological scoring. See verification.json for the local preparation checks.
These checks do not establish accuracy on a new study. The six-task growth benchmark belongs to a different endpoint and must not be attributed to this expression checkpoint. Opal's retained expression-composition comparisons include negative outcomes against selected simple references; the broader release also retains failed frozen external transfer and failed promotion criteria. The linked paper and scientific findings preserve the original scope.
Limitations
Population-mean effects do not reproduce cell-level distributions or identify causal molecular mechanisms. This model requires measured source-local single references; it is not a general cold-intervention or arbitrary-context predictor. Reliability and uncertainty diagnostics are exploratory, with no calibrated biological coverage guarantee. No clinical-use validation, biological discovery claim or frontier-level superiority is established.
Provenance and license
The source is Tahoe Bio's EmeraldBay dataset: https://huggingface.co/datasets/tahoebio/EmeraldBay . Pinned source revision: f2a0be6b02f731553657f0115c345b20bb020ede . Credit Tahoe Bio and the EmeraldBay creators listed in the pinned dataset card: https://huggingface.co/datasets/tahoebio/EmeraldBay/blob/f2a0be6b02f731553657f0115c345b20bb020ede/README.md .
The source is recorded as CC BY 4.0. Opal creates matched aggregate effects from the first six release shards and fits a contextual kernel; the retained pilot has eight registered contexts. These are adaptations. Dataset-derived content retains CC BY 4.0 attribution requirements. references.json preserves source revision, source hashes, intervention definitions, controls and preparation scope. No raw single-cell dataset is included.
LICENSE contains CC BY 4.0. Original Opal software and the inference example use MIT terms in CODE_LICENSE.txt; that software license does not replace the model/reference license.
Reproducibility and citation
- Original manifest SHA-256:
c3a1287c438f698acc3521c5a3c0495c040e426eb07e443d3b315cce087b3f81. - Opal source release:
v1.0.0. - Original model files and their hashes: provenance.json.
- Full folder checksums: SHA256SUMS.txt.
- Scientific methods and negative findings: research findings and preprint.
- Citation metadata: CITATION.cff. Cite the source study/dataset as well as Opal.
Prepared by The Tiny Intelligence Lab. Original checkpoint and manifest bytes are preserved; no new model training or biological evaluation was performed during packaging.