Profile Clusters Derived from BLOCKS Suggest a Simple Model of Column Evolution in Multiple Alignments of Protein Families

BLOSUM and PAM series of protein substitution matrices are popular tools for scoring protein pairwise and multiple alignments. For protein multiple alignments there exists another representation of an evolving column, based on a set of predefined frequency profile patterns. For conserved sites, these profile patterns represent stationary points in a 20-dimensional profile space. There are 20 such patterns. All of them were derived from the BLOCKS database by applying a special clusterization procedure to frequency profiles obtained from BLOCKS alignment columns. To understand the nature of these clusters, random protein sequences were generated where all columns were obtained from a single amino acid type by applying transition probabilities to it, and the same clusterization procedure was applied to the generated frequency profiles. Similar twenty clusters were obtained. This means that for conservative columns, all amino acids in that column are derived from a single ancestor amino acid by a substitution random process with standard transition probabilities. For non-conservative columns there are, generally, no regularities in the amino acid types present therein. Based on the COG database, a formula was obtained to distinguish between functionally important and unimportant columns. If the odds ratio of likelihoods of the most probable ancestral amino acid to the third most probable ancestral amino acid exceeds a critical value, then this column is predicted to be conservative and it is the result of evolution of the ancestral amino acid. Otherwise, this column is considered to be a “garbage” column. When building a consensus sequence from a multiple alignment, we can represent this column as a “garbage” symbol having zero value with any amino acid in a substitution matrix.