DeSpotX: Identifiability-Based Decontamination for Spatial Transcriptomics.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) offers gene expression profiling in native tissue context.
• Transcript contamination between adjacent cells in ST data compromises downstream analyses.
• Current decontamination methods lack spatial awareness and can be ambiguous.

Purpose of the Study:

• To develop a novel computational method for accurate transcript decontamination in spatial transcriptomics data.
• To improve the biological interpretability of spatial transcriptomics by resolving ambiguous contamination.
• To enhance marker-gene specificity and cell-cell communication network inference.

Main Methods:

• Introduced DeSpotX, a deep generative model utilizing anchor genes to constrain contamination decomposition.
• Employed spatial information for local contamination estimation via a cluster-masked, distance-weighted average.
• Incorporated a learned diffusion prior to prevent over-correction of low-expression signals.

Main Results:

• DeSpotX achieved AUROC > 0.94 across five simulated datasets and four ST platforms, outperforming baselines by 0.02-0.12.
• Demonstrated robustness to inaccuracies in cell-cluster annotation and anchor gene identification.
• Validated improved marker-gene specificity, spatial coherence, and biologically relevant cell-cell communication networks on real tissues.

Conclusions:

• DeSpotX effectively resolves contamination ambiguity in spatial transcriptomics data.
• The method enhances the accuracy and biological relevance of spatial transcriptomics analyses.
• Iterative refinement of decontamination and cell-cluster annotation further improves biological insights, such as ligand-receptor signaling localization.