---
license: cc-by-nc-4.0
library_name: terra-st
tags:
- spatial-transcriptomics
- foundation-model
- jepa
- single-cell
pipeline_tag: feature-extraction
---
# TERRA-96M

JEPA-based spatial-transcriptomics foundation model (TERRA).
Code & docs: https://github.com/Lotfollahi-lab/terra

## Training data
Trained on a 96M-cell subset of **HST-Corpus-112M**; the remaining cells are held out for benchmarking and downstream analyses. See the manuscript for details.

## Files
- `model_checkpoint.pt` — target-encoder weights (inference)
- `model_config.yaml` — model / tokenization config
- `token_dictionary.pkl` — gene-token vocabulary
- `ensembl_dictionary.pkl` — gene-name to Ensembl-ID mapping (harmonization)
- `gene_count_dictionary.pkl` — gene occurrence counts (rare-gene filtering)

## Usage
```python
from app.huggingface import download_pretrained
from app.inference import harmonize_tokenize_embed_pipeline

d = download_pretrained("Lotfollahi-lab/TERRA-96M")
adata = harmonize_tokenize_embed_pipeline(
    adata=adata,
    model_folder_path=d,            # gene-reference files auto-resolved from here
    # ... sample_key / batch_key / etc.
)
```

## Citation
<add paper / bioRxiv reference>