PHATE<T>

Models & Types ยท AiDotNet.Preprocessing.DimensionalityReduction

PHATE: Potential of Heat-diffusion for Affinity-based Transition Embedding.

Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> โ†’ PHATE<T>

For Beginners

PHATE excels at revealing data trajectories:

  • Captures smooth progression paths in data (differentiation, time courses)
  • Preserves both local clusters and global connectivity
  • The potential distance emphasizes transitions between states
  • Particularly effective for biological data

Use cases:

  • Single-cell RNA sequencing visualization
  • Developmental trajectories and cell differentiation
  • Time-series data with smooth transitions
  • Any data with underlying continuous processes

How It Works

PHATE is a dimensionality reduction method designed for visualizing high-dimensional biological data, particularly single-cell data. It captures both local and global structure by using diffusion-based distances and a special potential distance metric.

The algorithm:

  1. Compute local affinities using adaptive bandwidth Gaussian kernel
  2. Construct diffusion operator from affinities
  3. Diffuse for t steps to capture multi-scale structure
  4. Compute potential distance using log transform
  5. Embed using MDS on potential distances

Constructors

Constructor Summary
PHATE(int nComponents, int nNeighbors, int diffusionTime, double decay, double gamma, int? randomState, int[] columnIndices) Creates a new instance of PHATE.

Properties

Property Summary
Matrix<T> Embedding { get; } Gets the embedding result.
bool SupportsInverseTransform { get; } Gets whether this transformer supports inverse transformation.
int DiffusionTime { get; } Gets the diffusion time parameter.
int NComponents { get; } Gets the number of components (dimensions).
int NNeighbors { get; } Gets the number of neighbors.

Methods

Method Summary
string[] GetFeatureNamesOut(string[] inputFeatureNames) Gets the output feature names after transformation.