Multi-view data visualisation via manifold learning.

Area of Science:

• Computational biology
• Data science
• Machine learning

Background:

• Manifold learning methods like t-SNE, LLE, and ISOMAP reduce dimensionality for data visualization.
• Multi-view data, originating from the same samples, presents unique analytical challenges.
• Visualizing each data-view separately often limits comprehensive pattern identification.

Purpose of the Study:

• To extend existing manifold learning techniques for effective multi-view data dimensionality reduction and visualization.
• To develop a unified approach for analyzing and clustering multi-view datasets.
• To improve the identification of underlying patterns and structures within complex biological data.

Main Methods:

• Proposed extensions of Student's t-distributed SNE (t-SNE), Locally Linear Embedding (LLE), and Isometric Feature Mapping (ISOMAP) for multi-view data.
• Incorporation of multi-view manifold learning embeddings into the K-means clustering algorithm.
• Extensive comparative analysis of novel and existing multi-view manifold learning algorithms on synthetic and real-world datasets.

Main Results:

• The proposed multi-view extension of t-SNE, termed multi-SNE, demonstrated superior performance in dimensionality reduction and visualization.
• Multi-SNE provided more comprehensible sample projections compared to single-view approaches.
• The method accurately identified sample clusters when integrated with K-means clustering.

Conclusions:

• Multi-SNE offers a powerful and effective solution for unified clustering and visualization of multi-view data.
• The approach shows significant promise for analyzing challenging datasets, such as multi-omics single-cell data.
• Multi-SNE enhances the ability to visualize cell heterogeneity and identify cell types in biological tissues.