Bgoood/SpatialGT-Pretrained

SpatialGT Pretrained Model

Model Description

This is the pretrained checkpoint of SpatialGT (Spatial Graph Transformer), a graph transformer model designed for spatial transcriptomics data analysis.

SpatialGT leverages spatial context through neighbor-aware attention mechanisms for:

• 🗺️ Spatial context learning from large-scale spatial transcriptomics data
• 🧬 Gene expression reconstruction
• 🔬 Perturbation simulation

Model Details

• Architecture: Graph Transformer with spatial neighbor attention
• Parameters: ~600M
• Training Data: Large-scale spatial transcriptomics atlas
• Input: Gene expression vectors with spatial coordinates
• Output: Contextualized gene expression representations

Usage

import torch
from pretrain.model_spatialpt import SpatialNeighborTransformer
from pretrain.Config import Config

# Load configuration
config = Config()

# Initialize model
model = SpatialNeighborTransformer(config)

# Load pretrained weights
from safetensors.torch import load_file
state_dict = load_file("model.safetensors")
model.load_state_dict(state_dict)

model.eval()

Files

• model.safetensors: Model weights in safetensors format
• training_args.bin: Training arguments
• trainer_state.json: Training state information

Citation

If you use this model, please cite our paper (details to be added upon publication).

License

MIT License