This chapter focuses  on techniques for embedding general metric
  spaces into low-dimensional Euclidean spaces, and the application of
  these techniques to protein segments.
  
  The basic concept is simple: By embedding a set of protein sequences
  into Euclidean space, each protein is mapped to a specific point in
  Euclidean space, where similar proteins are mapped to close points.
  Thus, distinct families of proteins are mapped to separate clusters,
  allowing us to predict the biological function of a protein by
  observing the other members in its cluster.  This methodology thus
  converts the problem of finding a protein's function to a problem of
  pattern recognition.
  
  This approach must address three problems: (1) Constructing a
  metric space on the protein sequence space, (2) Finding the optimal
  embedding, and (3) Discovering statistical regularities in Euclidean
  space (clustering). The first two stages are described in this
  chapter.