Classification of protein sequences by their dipeptide composition

A simple approach to scan quickly a large protein sequence database for homology is described. The approach used is strictly dependent on the database organization. A database has been compiled in which protein sequences are grouped into families of closely related proteins, each family being characterized by its average dipeptide composition. A new entry in the database can be allocated in a family by comparing its dipeptide composition with the average dipeptide composition of the families.