SCR2ST

SCR²-ST is a unified framework that leverages single-cell prior knowledge to guide efficient data acquisition and accurate expression prediction.

Comparison between traditional ST sampling and our active sampling. Left: Traditional ST methods rely on fixed-grid sampling regardless of biological importance, leading to redundant measurements in similar regions and inefficient use of sequencing budgets. Right: Our proposed approach actively selects informative spots by incorporating single-cell prior knowledge, reducing redundancy while preserving biologically diverse regions.

Overview

This framework addresses the challenge of predicting gene expression from histology images in spatial transcriptomics. We propose a reinforcement learning-based active sampling strategy that intelligently selects informative spots for training by leveraging:

• Single-cell manifold coverage: Ensures sampled spots cover diverse cell states in the scRNA-seq reference
• Cell type diversity: Maximizes the entropy of cell type distribution in the selected samples
• Spatial uniformity: Encourages spatially dispersed sampling for better tissue coverage

The model also incorporates a retrieval-augmented module that enhances predictions by referencing similar expression patterns from the training set.

Installation

pip install -r requirements.txt

Usage

python main.py \
    --dataset HER2 \
    --root_path /path/to/st_data \
    --patch_root /path/to/patches \
    --sc_root /path/to/sc_embeddings \
    --total_ratio 0.5 \
    --max_epochs 100 \
    --batch_size 128 \
    --gpu 0

Project Structure

├── main.py           # Training script
├── model.py          # Model architectures
├── dataset.py        # Dataset class
├── rl_sampler.py     # RL-based sampler
├── reward.py         # Reward functions
├── cross_fold.py     # Cross-validation splits
├── eval_metric.py    # Evaluation metrics
└── requirements.txt  # Dependencies