STEAM: Spatial Transcriptomics Evaluation Algorithm and Metric for clustering performance.

Area of Science:

• Spatial omics
• Computational biology
• Genomics

Background:

• Spatial transcriptomics enables gene expression analysis within tissue context.
• Validating clustering in spatial omics is challenging due to the absence of ground truth labels.
• A robust computational framework is needed to assess clustering performance and reproducibility.

Purpose of the Study:

• To introduce STEAM (Spatial Transcriptomics Evaluation Algorithm and Metric), a computational pipeline for evaluating spatial omics clustering.
• To assess the consistency and reliability of clustering results by preserving spatial proximity and gene expression patterns.
• To provide a tool for benchmarking different clustering approaches in spatial biology.

Main Methods:

• STEAM utilizes machine learning classification and prediction methods.
• The pipeline supports multi-sample training for cross-replicate consistency assessment.
• It evaluates clustering performance using metrics like Kappa score, F1 score, accuracy, and adjusted rand index.

Main Results:

• STEAM demonstrated robustness and generalizability across diverse spatial omics datasets (multi-cell to single-cell, transcriptomics, proteomics).
• The framework successfully compared spatial-aware and spatial-ignorant clustering methods.
• Results confirmed STEAM's utility in assessing clustering robustness and reproducibility.

Conclusions:

• STEAM offers a user-friendly and reliable tool for evaluating clustering in spatial omics data.
• It aids researchers in driving reproducible discoveries in spatial biology.
• The R software tool and source code are publicly available.