Dimensionality reduction by UMAP reinforces sample heterogeneity analysis in bulk transcriptomic data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transcriptomic analysis is crucial in biomedical research.
• Linear methods like Principal Component Analysis (PCA) are common for heterogeneity detection.
• Non-linear methods like t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) are effective for single-cell RNA sequencing.

Purpose of the Study:

• To evaluate the application and performance of non-linear dimensionality reduction methods (t-SNE, UMAP) in bulk transcriptomic analysis.
• To compare UMAP and t-SNE against conventional methods (PCA, Multidimensional Scaling [MDS]).
• To assess the ability of these methods to identify sample heterogeneity and biological patterns in large datasets.

Main Methods:

• Comparative analysis of four dimensionality reduction techniques: PCA, MDS, t-SNE, and UMAP.
• Application of these methods to 71 large-scale bulk transcriptomic datasets.
• Evaluation of performance based on differentiating batch effects, identifying biological groups, and revealing clusters.

Main Results:

• UMAP demonstrated superior performance compared to PCA and MDS.
• UMAP showed advantages over t-SNE in distinguishing batch effects and identifying biological groups.
• UMAP effectively revealed in-depth sample clusters with significant biological and clinical relevance in 2D space.

Conclusions:

• UMAP is a highly effective tool for visualizing and analyzing bulk transcriptomic data.
• UMAP enhances the identification of sample heterogeneity and biological patterns.
• The study recommends UMAP for large-scale bulk transcriptomic dataset analysis.