Fundamentals of Bioinformatics - Ogurtsov A.N. 2013

Methods of Bioinformatic Analysis
Substitution Matrices
BLOSUM Matrices

Another family of amino acid Substitution Matrices, the BLOSUM matrices, was developed in 1992 by Steven and Jorja Henikoff (Steven Henikoff, Jorja G. Henikoff) using the BLOCKS database of aligned protein sequences, hence the name BLOSUM (BLOcks SUbstitution Matrix).

METABOLISM/2.html">THE CONCEPT OF a "block" is derived from the notion of a "motif"—a conserved segment of an Amino Acid Sequence that confers a specific Structure or function to a protein (see [6], sec. 6.1). If motifs from a protein family are aligned without introducing gaps into the sequence, such a "motif of motifs" is called a block.

Examining contiguous combinations of Amino Acids found in each family of related Proteins aligned without gaps, the Henikoff couple calculated The ratio of observed amino acid residue substitution pairs to the expected number of substitutions for all considered related blocks at each sequence position.

To reduce the overwhelming contribution of substitutions occurring in the most closely related sequences, such sequences were pre-filtered, and their substitution counts were averaged; it is these weighted average values that were used to represent the entire group of related proteins.

The blocks were then grouped by identity, and a substitution matrix was calculated within each group: for blocks with 45% identity, the BLOSUM45 matrix; for 62% similarity, the BLOSUM62 matrix; for 80% similarity, the BLOSUM80 matrix, and so on.

When aligning Amino acid sequences using a BLOSUM matrix, the ClustalW program suggests default penalties of d = 10 for gap opening (gap open) and e = 0.1 for gap extension (gap extension).

Like PAM Matrices, BLOSUM matrices are also based on the concept of mutation frequencies. However, when deriving BLOSUM matrices, mutation frequencies are determined by searching the BLOCKS database, and the numbers assigned to BLOSUM matrices do not carry the same interpretation as in the case of PAM matrices. Moreover, any potential bias introduced by counting multiple contributions from identical residue pairs is eliminated by grouping sequence segments based on their minimum percent identity.

When constructing BLOSUM matrices, groups of similar sequences are effectively treated as separate (weighted average) sequences. Protein blocks contain local multiple alignments of distantly related sequences (as opposed to the closely related sequences used in PAM).

As with PAM matrices, the results in BLOSUM matrices are presented as log-odds scores of substitution frequencies. To avoid fractional numbers, all matrix values are also multiplied by 10 and rounded.

Tables 12-15 present the BLOSUM45, BLOSUM50, BLOSUM62, and BLOSUM80 amino acid substitution matrices, which are most frequently used in alignment calculations.

In most alignment programs, the BLOSUM62 matrix is used by default.

Class="center">Table 12 - BLOSUM45 amino acid substitution matrix

Table 13 - BLOSUM50 amino acid substitution matrix

Table 14 - BLOSUM62 amino acid substitution matrix

Table 15 - BLOSUM80 amino acid substitution matrix

Generally speaking, a BLOSUM matrix represents an evolutionary model in matrix form, as it is derived from direct data rather than extrapolated values, unlike PAM matrices.