liveBook · Manning

Best matching unit

As explained in Build AI Drug Discovery Pipelines

A best matching unit is the node in a self-organizing map whose weight vector is closest to a given input vector.

A best matching unit (BMU) is the self-organizing-map node whose weight vector is closest to an input vector. For each input, the map compares distances to every node and selects the closest winner. section 7.1.2

A BMU participates in the SOM training loop:

  • Represent an input, such as an RGB color or molecular fingerprint, as a vector. section 7.1.2
  • Calculate its distance from every node and choose the node with the smallest distance. section 7.1.2
  • Update the BMU and neighboring nodes toward the input, with the neighborhood function controlling update strength. section 7.1.2

Compound mapping

During molecular screening, a compound’s fingerprint selects a BMU on a two-dimensional grid. Repeated assignments and updates place similar compounds near one another, allowing nodes to show enrichment for active or inactive compounds.

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains vector-distance selection and iterative neighbor updates, including Gaussian weighting, but stops short of deriving SOM convergence.
Examples
Uses colored RGB beads, molecular fingerprints, active or inactive compound classes, and two-dimensional SOM grids.
What it assumes
Assumes vector representations and distance calculations while introducing BMU selection within the iterative SOM procedure.
What you can do after
Apply it via MiniSom to identify winning nodes for fingerprint vectors and inspect compound organization on a trained map.
How it connects
Connects self-organizing maps with weight vectors, Euclidean distance, neighborhood functions, learning rates, and molecular fingerprints.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.