Table 4 ROC scores for the SW algorithm and the LA kernel in the independent dataset. The first column shows the scoring method. For example, BLOSUM62SWOPT is the matrix optimized for the SW algorithm starting from the BLOSUM62. The second column shows the performance of each score matrix by the SW algorithm on the COG distant test set. The following columns show the performance, in terms of average ROC score, of each matrix used in combination with either the SW algorithm or the LA kernel on four different datasets. The second row shows the performance of PSI-BLAST with the BLOSUM62 with gap open and extension parameters set to 11 and 1 (default), respectively. The best ROC score in each dataset is highlighted in bold font.

From: Optimizing amino acid substitution matrices with a local alignment kernel