t-SNE
Definition
t-SNE (t-Distributed Stochastic Neighbor Embedding) is a non-linear dimensionality reduction technique widely used in bioinformatics to visualize high-dimensional biological data in two or three dimensions. It preserves local structure by converting similarities between data points into joint probabilities and minimizing the Kullback-Leibler divergence between these distributions in high and low dimensions. In life sciences, t-SNE is particularly valuable for visualizing single-cell RNA sequencing data, revealing distinct cell populations, identifying rare cell types, and exploring complex molecular phenotypes. Unlike PCA, t-SNE excels at capturing non-linear relationships and clustering patterns, making it essential for exploratory analysis of complex biological datasets where traditional linear methods fail to reveal meaningful structure.
Visualize t-SNE in Nodes Bio
Researchers can use Nodes Bio to transform t-SNE cluster results into interactive network graphs where nodes represent individual cells or samples, and edges connect similar entities based on t-SNE proximity. This enables exploration of relationships between cell populations, identification of transition states, and integration of pathway enrichment data. Users can overlay gene expression patterns, protein markers, or clinical metadata onto network structures derived from t-SNE coordinates.
Visualization Ideas:
- Cell population networks from single-cell t-SNE clusters with trajectory inference overlays
- Sample similarity networks where t-SNE coordinates determine edge weights between patients or conditions
- Multi-omics integration networks combining t-SNE results from transcriptomics, proteomics, and metabolomics data
Example Use Case
A cancer immunology team performs single-cell RNA-seq on tumor-infiltrating lymphocytes from melanoma patients. After applying t-SNE, they identify seven distinct T-cell clusters. By importing these results into Nodes Bio, they create a network where each cluster becomes a node, with edges weighted by transcriptional similarity. They overlay checkpoint inhibitor response data and discover that a rare CD8+ exhausted T-cell population, positioned between two major clusters in t-SNE space, correlates with treatment resistance, revealing a potential therapeutic target.