What is it about?
This paper systematically compares widely used dimensionality reduction methods (such as PCA, non-negative matrix factorization (NMF), and deep learning–based autoencoders) for analyzing spatial transcriptomics data. Using a large cholangiocarcinoma tissue dataset, the authors evaluate how different techniques influence downstream clustering, biological interpretation, and marker-gene fidelity. In addition to benchmarking existing methods, the study introduces two biologically motivated evaluation metrics (1) Cluster Marker Coherence (CMC) and (2) Marker Exclusion Rate (MER), and a simple post-processing strategy that reassigns cells to more biologically appropriate clusters based on marker expression.
Featured Image
Photo by ANIRUDH on Unsplash
Why is it important?
Dimensionality reduction is a foundational step in nearly all spatial transcriptomics pipelines, yet method choice is often arbitrary or driven by convenience. This work shows that different methods can lead to meaningfully different biological conclusions, even when applied to the same data. By providing a unified, quantitative benchmarking framework and introducing interpretable biological metrics, the study enables researchers to make principled, application-specific choices rather than relying on default tools. The proposed MER-guided reassignment further demonstrates that biological accuracy can be improved without redesigning entire pipelines, making the findings immediately actionable for the community.
Perspectives
For computational biologists: The study highlights that no single dimensionality reduction method is universally optimal; method selection should depend on whether geometric separation, reconstruction accuracy, or biological marker fidelity is the primary goal. For experimental and translational researchers: The results emphasize that clustering artifacts can arise purely from computational choices, underscoring the need for biologically grounded validation metrics when interpreting spatial domains or cell types. For method developers: The work opens avenues for integrating marker-aware objectives directly into dimensionality reduction and clustering algorithms, particularly for spatially aware models.
Tania Banerjee
University of Houston
Read the Original
This page is a summary of: Benchmarking Dimensionality Reduction Techniques for Spatial Transcriptomics, October 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3765612.3767237.
You can read the full text:
Contributors
The following have contributed to this page
