GAADE: identification spatially variable genes based on adaptive graph attention network

Abstract

The rapid advancement of spatial transcriptomics (ST) sequencing technology has made it possible to capture gene expression with spatial coordinate information at the cellular level. Although many methods in ST data analysis can detect spatially variable genes (SVGs), these methods often fail to identify genes with explicit spatial expression patterns due to the lack of consideration for spatial domains. Considering spatial domains is crucial for identifying SVGs as it focuses the analysis of gene expression changes on biologically relevant regions, aiding in the more accurate identification of SVGs associated with specific cell types. Existing methods for identifying SVGs based on spatial domains predefine spot similarity before training, which prevents adaptive learning and limits generalizability across different tissues or samples. This limitation may also lead to inaccurate identification of specific genes at boundary regions. To address these issues, we present GAADE, an unsupervised neural network architecture based on graph-structured data representation learning. GAADE stacks encoder/decoder layers and integrates a self-attention mechanism to reconstruct node attributes and graph structure, effectively capturing spatial domain structures of different sections. Consequently, we confine the identification of SVGs within spatial domains. By performing differential expression analysis on spots within the target spatial domain and their multi-order neighbors, GAADE detects genes with enriched expression patterns within defined domains. Comparative evaluations with five other popular methods on ST datasets across four different species, regions and tissues demonstrate that GAADE exhibits superior performance in detecting SVGs and capturing the extent of spatial gene expression variation.

Keywords: ST-seq; graph attention auto-encoders; spatial domain; spatial neighbor graph; spatially variable gene.

Figure 1

Workflow of GAADE.

Figure 2

Identification of spatial domains in the LIBD dorsolateral prefrontal cortex data by GAADE. (a) Boxplot showing the clustering accuracy of GAADE and current state-of-the-art algorithms (SpaGCN, DeepST, stLearn, SEDR, and SCANPY) across all 12 sections of the DLPFC dataset, as measured by the ARI scores. (b) on slice 151,673, the actual locations of points are mapped to their spatial positions, which are divided into six cortical layers (L1-L6) and white matter (WM). (c) Clustering assignments generated by GAADE, SCANPY, SpaGCN, DeepST, stLearn, and SEDR in the DLPFC slice 151,673.

Figure 3

Performance of SVGs detected by GAADE in the LIBD dorsolateral prefrontal cortex data. (a) Spatial expression patterns of SVGs in spatial domains 1 (MOBP) and 5 (PCP4) in slice 151,673. (b) Boxplot of Moran’s I and Geary’s C values for SVGs detected by GAADE, SpaGCN, Squidpy, SpatialDE, ScGCO, and SPARK in slice 151,673. (c) High expression genes identified by the SpaGCN method in spatial domain 5 of the DLPFC 151673 slice data.

Figure 4

Spatial domains generated by GAADE in the LIBD dorsolateral prefrontal cortex data before and after the introduction of the attention mechanism.

Figure 5

Spatial domains and SVGs detected in mouse coronal brain slice data. (a) Clustering results of GAADE for mouse coronal brain slices. (b) Boxplot of Moran’s I and Geary’s C values for SVGs detected by GAADE, SpaGCN, Squidpy, SpatialDE, ScGCO, and SPARK in mouse coronal brain slice data. (c) Spatial expression pattern of SVGs in mouse coronal brain spatial domain 20 (Camk2a). (d) High expression genes identified by the SpaGCN method in spatial domains 12 and 5 of mouse coronal brain slice data.

Figure 6

Analysis of SVGs detected in ductal carcinoma in situ slice data of human breast tissue. (a) Boxplot of Moran’s I and Geary’s C values for SVGs detected by GAADE, SpaGCN, Squidpy, SpatialDE, ScGCO, and SPARK in ductal carcinoma in situ slice data of human breast tissue. (b) Spatial expression patterns of SVGs in spatial domains 6 and 8 of ductal carcinoma in situ slices of human breast tissue (In the lower right corner of Figure 6, the outer annulus region is distinctly annotated as spatial domain 8, whereas the inner annulus area is precisely labeled as spatial domain 6). (c) Heatmap summarizing the high-expression genes identified by the SpaGCN method in spatial domain 6.

Figure 7

Spatial domains and SVGs detected in the anterior sagittal slice data of mouse brain tissue. (a) Boxplot of Moran’s I and Geary’s C values for SVGs detected by GAADE, SpaGCN, Squidpy, SpatialDE, ScGCO, and SPARK in anterior sagittal slice data of mouse brain tissue. (b) High expression genes identified by the SpaGCN method in spatial domains 10 and 8 of mouse sagittal brain slice data. (c) Clustering results of GAADE for the anterior brain tissue slices. (d) Spatial expression patterns of SVGs in spatial domains 17 and 18 of the anterior sagittal slices of mouse brain tissue (In the central region of the figure, the upper half is precisely annotated to signify Spatial Domain 18, while the lower half is distinctly labeled to denote Spatial Domain 17).