In single-cell RNA sequencing analysis, what is the primary objective of applying manifold learning techniques like UMAP or t-SNE to high-dimensional gene expression matrices?

Lara Emad Aljudaibi

“The primary objective of applying manifold learning techniques like UMAP or t-SNE in a single-cell RNA sequencing is dimensionality reduction, which allows researchers to visualize and interpret high-dimensional gene expression data in a human-readable two-dimensional or three-dimensional space. A single-cell RNA sequencing matrix containd thousands of variables, where each variable represents the expression level of a single gene across thousands of individual cells, creating a high-dimensional space that is impossible for humans to visualize directly. Manifold learning assumes that while data exist in thousands of dimensions, the actual biological signal lies on a lower-dimensional structure or surface, known as a manifold, hidden within that space. These algorithms work by calculating the similarity between cells based on their overall gene expression profiles and then projecting those cells onto a lower-dimensional plane while attempting to preserve the local neighborhood structure of the data. By maintaining these local relationships, cells with similar expression patterns— such as those belonging to the same cell type or developmental state— are grouped together into clusters, while distinct cell populations are pushed apart. This transformation makes it possible to identify cell types, discover rare cell populations, and visualize the progressions of biological processes like differentiation, which would remain obscured in the raw, high-dimensional numerical matrix.”

94.0%