CPS: Mapping Physical Coordinates to High-Fidelity Spatial Transcriptomics via Privileged Multi-Scale Context Distillation

📖 Overview

We introduce the Cell Positioning System (CPS), a context-aware implicit neural representation framework designed to map physical coordinates to high-fidelity spatial transcriptomics via a privileged multi-scale context distillation strategy. CPS treats multi-scale tissue niches as privileged information, employing a teacher network equipped with a multi-scale niche attention mechanism to capture adaptive biological interactions during training. This structural knowledge is explicitly distilled into a student coordinate network, enabling the generation of context-aware expression landscapes solely from spatial coordinates during inference.

🚀 Reproduction of results

We provide comprehensive Jupyter notebooks to reproduce the results and figures presented in the paper. The experiments are organized into three main directories: benchmark, case_study, and Interpret.

• 📂 benchmark/: Contains benchmarks for Spatial Imputation and Gene Imputation on the 12-slice DLPFC dataset.

  • 1_spatiual_imputation_DLPFC.ipynb: Benchmarking spatial imputation performance on #151673 .
  • 2_gene_imputation_DLPFC.ipynb: Benchmarking gene imputation performance on #151673.
  • 3_DLPFC_12_SI.ipynb & 4_DLPFC_12_GI.ipynb: Detailed benchmarking scripts for the 12 DLPFC slices.

• 📂 case_study/: Demonstrates Super-Resolution (SR) capabilities and Scalability analysis.

  • 1_MBSP_SR.ipynb: Super-Resolution Task. Reconstructs high-fidelity gene expression at arbitray resolution using MBSP data.
  • 2_HD_Scalable.ipynb: Scalability on Visium HD. Demonstrates efficient processing of Visium HD data.
  • 3_MED_Efficient_*.ipynb: Efficient on Mouse Embryo. Validates performance and efficiency across varying data scales using the Mouse Developing Embryo (MED) atlas.
  • time_compute.ipynb: Analysis of computational time and resource usage.

• 📂 Interpret/: Focuses on model interpretability.

  • 1_HBC_interpret_attn_scores.ipynb: Interpretability Analysis. Visualizes multi-scale attention scores on the Human Breast Cancer (HBC) dataset to decode tissue heterogeneity.

• To run these notebooks, ensure you have installed the required dependencies and downloaded the necessary datasets.

📈 SRT data for evaluating CPS

The datasets analyzed in this study are publicly available from their respective repositories:

• The human Dorsolateral Prefrontal Cortex (DLPFC) dataset is available via the spatialLIBD package or at http://research.libd.org/spatialLIBD/.
• The Human Breast Cancer, Mouse Posterior Brain (Visium) and Mouse Brain Visium HD datasets can be downloaded from the 10x Genomics website https://www.10xgenomics.com/resources/datasets.
• The Stereo-seq Mouse Embryo Atlas is accessible through the MOSTA database https://db.cngb.org/stomics/mosta/.

🤝 Software depdendencies

• scanpy==1.9.8
• scikit-learn==1.3.2
• scipy==1.10.1
• squidpy==1.2.3
• torch==2.1.0+cu121
• torch_geometric==2.5.3
• transformers==4.46.3

---