stPlus

stPlus enhances spatial transcriptomic data by leveraging single-cell RNA sequencing (scRNA-seq) as a reference to improve spatial gene expression prediction and characterization of cellular heterogeneity.

Key Features:

• Reference-Based Enhancement: stPlus utilizes scRNA-seq data as a reference to improve the accuracy of spatial gene expression predictions in spatial transcriptomics datasets.
• Auto-Encoder Architecture with Tailored Loss Function: It employs an auto-encoder neural network optimized with a tailored loss function to perform joint embedding of scRNA-seq and spatial transcriptomics data.
• Weighted k-Nearest-Neighbor (k-NN) Prediction: Spatial gene expressions are predicted using a weighted k-nearest-neighbor algorithm applied in the joint embedding space.
• Performance Metrics: Performance is quantified using gene-wise and cell-wise Spearman correlation coefficients.
• Clustering-Based Performance Assessment: A clustering-based approach is used to evaluate enhancement by assessing identification of cell populations in enhanced versus raw measured data.
• Robust and Scalable: The method is robust across varying gene detection sensitivity, sample sizes, and numbers of spatially measured genes.

Scientific Applications:

• Cell Population Identification: Enhances resolution and accuracy of spatial transcriptomics data to enable more precise identification and characterization of cell populations within tissue.
• Characterization of Spatial Cell Heterogeneity: Enables characterization of spatial heterogeneity by predicting gene expression present in scRNA-seq but not directly measured in imaging-based spatial transcriptomic datasets.

Methodology:

stPlus integrates scRNA-seq and spatial transcriptomics via joint embedding using an auto-encoder optimized with a tailored loss function, followed by weighted k-nearest-neighbor prediction of spatial gene expression and evaluation using clustering-based assessment and gene-wise and cell-wise Spearman correlation coefficients.