DNA Scoring Matrices: Application to Detect Divergent Repeats

Abstract

The log-odd scoring matrices for proteins (PAM series) were first published nearly 20 years ago. We describe scoring matrices for DNA that are based on the frequencies of nucleotide substitutions and insertions/deletions observed in the evolution of repeats. Series of scoring matrices with different relative entropies for several repeat families were calculated. These matrices are asymmetric; for example, c→t has a different score from t→c; insertions occur much less frequently (and have lower scores) than deletions. We calculated separate scores for c and g when they form a eg dinucleotide. The scoring matrices with different relative entropies describe nearly neutral DNA evolution at various evolutionary distances. We used scoring matrices derived from mammalian interspersed repeats (MIRs) to search for copies of this repeat in GenBank. The sensitivity of both Smith-Waterman and BLAST searches improves dramatically when we use scoring matrices instead of default match/mismatch scores. These matrices can become new defaults for DNA similarity searches in popular BLASTN and FASTA programs. The DNA scoring matrices can be used in all similarity searches that require high sensitivity to detect distantly related nucleotide sequences that evolve under weak selective constraints, for example, repeats, pseudogenes, introns, and other noncoding sequences.