Reports in Informatics Relation Patterns and Their Automatic Discovery in Biosequences Relation Patterns and Their Automatic Discovery in Biosequences

We have extended the pattern language used in PROSITE to enable it to describe dependencies between amino acid residues. We have developed a minimum description length principle based tness measure evaluating the signiicance of such patterns in relation to a set of sequences, and an algorithm automatically nding signiicant patterns in unaligned sequences. Computing experiments are reported showing that the method is able to discover residue dependences in many PROSITE families, as well as some reported earlier by another author as possibly resulting from structural constraints. The method has also demonstrated that many seemingly correlated amino acid positions may be resulting from diierent subfamilies being present in the family.