Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

stMCDI: Masked Conditional Diffusion Model with Graph Neural Network for Spatial Transcriptomics Data Imputation

Xiaoyu Li, Wenwen Min, Shunfang Wang, Changmiao Wang, Taosheng Xu

TL;DR

This work tackles the problem of extensive missing values in high-resolution spatial transcriptomics by leveraging spot spatial coordinates through a graph neural encoder and a masked self-supervised training regime. It introduces stMCDI, a masked conditional diffusion model that uses unmasked data as a priori conditioning and a cross-attention enhanced UNet to impute missing gene expressions while preserving the data distribution. The approach achieves state-of-the-art performance across six real ST datasets against fourteen baselines, with ablations confirming the value of the GNN encoder, masking strategy, and conditioning mechanism. The results highlight the practical potential of combining graph-based spatial encoding with conditional diffusion for accurate, distribution-preserving imputation in spatial omics, and point to future directions including multi-modal integration and downstream analysis improvements.

Abstract

Spatially resolved transcriptomics represents a significant advancement in single-cell analysis by offering both gene expression data and their corresponding physical locations. However, this high degree of spatial resolution entails a drawback, as the resulting spatial transcriptomic data at the cellular level is notably plagued by a high incidence of missing values. Furthermore, most existing imputation methods either overlook the spatial information between spots or compromise the overall gene expression data distribution. To address these challenges, our primary focus is on effectively utilizing the spatial location information within spatial transcriptomic data to impute missing values, while preserving the overall data distribution. We introduce \textbf{stMCDI}, a novel conditional diffusion model for spatial transcriptomics data imputation, which employs a denoising network trained using randomly masked data portions as guidance, with the unmasked data serving as conditions. Additionally, it utilizes a GNN encoder to integrate the spatial position information, thereby enhancing model performance. The results obtained from spatial transcriptomics datasets elucidate the performance of our methods relative to existing approaches.

stMCDI: Masked Conditional Diffusion Model with Graph Neural Network for Spatial Transcriptomics Data Imputation

TL;DR

This work tackles the problem of extensive missing values in high-resolution spatial transcriptomics by leveraging spot spatial coordinates through a graph neural encoder and a masked self-supervised training regime. It introduces stMCDI, a masked conditional diffusion model that uses unmasked data as a priori conditioning and a cross-attention enhanced UNet to impute missing gene expressions while preserving the data distribution. The approach achieves state-of-the-art performance across six real ST datasets against fourteen baselines, with ablations confirming the value of the GNN encoder, masking strategy, and conditioning mechanism. The results highlight the practical potential of combining graph-based spatial encoding with conditional diffusion for accurate, distribution-preserving imputation in spatial omics, and point to future directions including multi-modal integration and downstream analysis improvements.

Abstract

Spatially resolved transcriptomics represents a significant advancement in single-cell analysis by offering both gene expression data and their corresponding physical locations. However, this high degree of spatial resolution entails a drawback, as the resulting spatial transcriptomic data at the cellular level is notably plagued by a high incidence of missing values. Furthermore, most existing imputation methods either overlook the spatial information between spots or compromise the overall gene expression data distribution. To address these challenges, our primary focus is on effectively utilizing the spatial location information within spatial transcriptomic data to impute missing values, while preserving the overall data distribution. We introduce \textbf{stMCDI}, a novel conditional diffusion model for spatial transcriptomics data imputation, which employs a denoising network trained using randomly masked data portions as guidance, with the unmasked data serving as conditions. Additionally, it utilizes a GNN encoder to integrate the spatial position information, thereby enhancing model performance. The results obtained from spatial transcriptomics datasets elucidate the performance of our methods relative to existing approaches.
Paper Structure (41 sections, 42 equations, 5 figures, 4 tables, 2 algorithms)

This paper contains 41 sections, 42 equations, 5 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Self-supervised Learning for Data Imputation
  4. Imputation Data via Generative Model
  5. Spatial Transcriptomics Data Imputation
  6. Proposed Method: stMCDI
  7. Problem formulation
  8. Mask and re-mask strategy in stMCDI
  9. GNN Encoder in stMCDI for integrating ST location information
  10. Constructing adjacency matrices
  11. Constructing latent representation with GCN
  12. Conditional score-based diffusion model in stMCDI
  13. Denoising diffusion probabilistic model and Score-based diffusion model
  14. Conditioning mechanisms in stMCDI
  15. Imputation with stMCDI
  16. ...and 26 more sections

Figures (5)

  • Figure 1: The network architecture of the proposed stMCDI model. Our model input has two parts: spot gene expression matrix and spot spatial location information. Build a graph based on the location information of each adjacent spot. Then the diffusion model is used to restore the masked representation to achieve the purpose of imputation.
  • Figure 2: Visualization of the imputation performance of various baseline methods. In this figure, we only show several baselines: CSDI, STAGATE, gimVI, scGNN, DCA, KNN, Mean, etc. For the remaining Baslines, please refer to Appendix F.2.
  • Figure 3: Different mask ratio of imputation performance in four metrics. We adopt different mask proportions for each sample, and we find that the performance of stMCDI reaches its best when the mask proportion is around 60%.
  • Figure 4: Visualization of the imputation performance of various generative methods across six distinct spatial transcriptomic datasets.
  • Figure 5: Visualization of the imputation performance of various baseline methods across six distinct spatial transcriptomic datasets.