TMAP

TMAP represents high-dimensional datasets as minimum spanning trees to visualize and explore intrinsic relationships in large-scale scientific data.

Key Features:

• Locality Sensitive Hashing: Employs locality sensitive hashing to efficiently handle large datasets while preserving local structure.
• Kruskal’s Minimum-Spanning Tree Algorithm: Constructs a minimum spanning tree using Kruskal’s algorithm to represent intrinsic relationships by minimizing total edge weight.
• Multilevel Multipole-Based Graph Layout Algorithm: Applies a multilevel multipole-based graph layout algorithm to organize the tree into a hierarchical, interpretable layout.

Scientific Applications:

• Broad domains: Cheminformatics, genomics, literature mining, cancer biology, and particle physics.
• ChEMBL — 1.16 million drug-like molecules: Visualizes large-scale chemical space to facilitate analysis of chemical properties and potential interactions.
• FDB17 — 10.1 million small molecule fragments: Enables exploration of fragment-based drug discovery data.
• PDB Databank — 131 thousand 3D structures of biomolecules: Assists structural biology research by mapping 3D biomolecular structures.
• GUTENBERG Data Set: Enables exploration of large literary datasets.
• PANSCAN Data Set: Supports cancer genomics studies.
• MiniBooNE Data Set: Facilitates visualization of particle physics data.

Methodology:

TMAP uses locality sensitive hashing to handle large, high-dimensional data while preserving local structure, constructs a minimum spanning tree via Kruskal’s algorithm, and applies a multilevel multipole-based graph layout; this approach reduces memory requirements and computational time compared to t-SNE or UMAP and produces an explorable tree representation.