EnDecon: cell type deconvolution of spatially resolved transcriptomics data via ensemble learning

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (SRT) provides tissue architecture insights but often lacks single-cell resolution.
• SRT data (spots) contain mixed cell types, necessitating deconvolution methods.
• Existing deconvolution methods vary in accuracy due to different modeling strategies.

Purpose of the Study:

• To develop an advanced cell-type deconvolution method for SRT data.
• To improve the accuracy of predicting cell-type compositions in spatial transcriptomics.
• To leverage and integrate existing deconvolution methods for enhanced performance.

Main Methods:

• Introduced EnDecon, a weighted ensemble learning deconvolution method.
• Integrated multiple base deconvolution results using a weighted optimization model.
• Validated performance through simulation studies and application to real SRT datasets.

Main Results:

• EnDecon demonstrated superior performance compared to existing methods in simulations.
• Learned weights in EnDecon correlated positively with base deconvolution method performance.
• Applied to real data, EnDecon identified cell types, localized them spatially, and revealed enrichment patterns.

Conclusions:

• EnDecon offers a more accurate approach for cell-type deconvolution in SRT.
• The method provides valuable insights into tissue spatial heterogeneity and regionalization.
• EnDecon enhances the interpretation of spatial transcriptomics data.