Previous algorithms for motif discovery and protein alignment have used a variety of scoring functions, each specialized to find certain types of similarity in preference to others. Here we present a novel scoring function that combines the relative entropy score with a sensitivity to amino acid similarities, producing a score that is highly sensitive to the types of weakly-conserved patterns that are typically seen in proteins. We investigate the performance of the hybrid score compared to existing scoring functions. We conclude that the hybrid is more sensitive than previous protein scoring functions, both in the initial detection of a weakly conserved region of similarity, and given such a similarity, in the detection of weakly-conserved instances.
展开▼