Imputing spatial transcriptomics through gene network constructed from protein language model.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Image-based spatial transcriptomics offers single-cell resolution but limited gene detection.
• Existing imputation methods struggle with accuracy and cell population identification due to ignoring gene relationships.

Purpose of the Study:

• To introduce stImpute, a novel computational method for imputing gene expression in spatial transcriptomics data.
• To leverage reference single-cell RNA sequencing (scRNA-seq) data and gene networks for improved imputation.

Main Methods:

• stImpute utilizes an autoencoder for gene expression embedding and a graph neural network (GNN) for imputation.
• Gene networks are constructed using embeddings from the ESM-2 protein language model, capturing gene-gene relationships.
• Nearest neighboring cells between spatial and scRNA-seq datasets are identified to guide imputation.

Main Results:

• stImpute demonstrates superior performance in imputation and cell clustering compared to existing state-of-the-art methods across multiple datasets.
• The method provides robust and consistent results, insensitive to parameter variations.
• A deep learning model quantifies gene prediction uncertainty.

Conclusions:

• stImpute effectively addresses limitations in spatial transcriptomics imputation by incorporating gene networks.
• The approach enhances the accuracy of gene expression data and cell population identification.
• stImpute offers a robust and reliable tool for spatial transcriptomics analysis.