Substitution Matrices in Sequence Comparisons

The concept of using a substitution matrix, i.e. a knowledge-derived table for judging amino acid similarities, was introduced by Dayhoff et al. in 19781171. Most types of currently used substitution matrices are derived from the analysis of well established alignments, by counting which types of residues are frequently substi- 

The left half of the figure shows an unreasonable alignment resulting from too low deletion/insertion penalties. The right half shows a better alignment, although the % identity score is worse. 

All commonly used substitution matrices are derived from a large collection of protein alignments, containing both enzymes and non-enzymes. Thus, favorable residue groupings tend to refled a structural compatibility rather than a functional equivalence. It would be expected that a substitution matrix derived from particular sets of enzymes would have quite different values for residues that are frequently found in active sites. 

---