"... In PAGE 15: ...ttp://monod.uwaterloo.ca/supplements/03seeds. We used the training samples to search for the optimal seed sets with our LP algorithm (note that no probabilistic model is used here), and then tested our seed sets on the test samples. [ Table6 about here.] Table 6 shows the sensitivity of the seed sets with weight 10 and weight 11.... In PAGE 15: ... [Table 6 about here.] Table6 shows the sensitivity of the seed sets with weight 10 and weight 11. As shown in these two figures, four seeds with weight 11 are 2% more sensitive than a single seed with weight 10.... ..."

"... In PAGE 12: ....943 0.797 2.632 0.997 0.943 10.304 0.958 0.814 2.982 0.998 0.946 11.341 training samples to search for the optimal seed sets with our LP algorithm (note that no probabilistic model is used here), and then tested our seed sets on the test samples. Table6 shows the sensitivity of the seed sets with weight 10 and weight 11. As shown in these two figures, four seeds with weight 11 are 2% more sensitive than a single seed with weight 10.... ..."

"... In PAGE 6: ... Because human chromosome 20 is considered to be completely ortholgous to mouse chromosome 2, very little difference should be seen between the coverage of human chromosome 20 by mouse chromosome 2 and by the whole mouse genome. The results are summarized in Table1 . The overall conclusion is that while SLAGAN may not be quite as sensitive as BLASTZ on the whole genome scale, it is able to align a larger percentage of the orthologous regions of human chromosome 20 and has higher specificity as shown by the lower percentage of non-orthologous alignments.... ..."

"... In PAGE 6: ... Because human chromosome 20 is considered to be completely ortholgous to mouse chromosome 2, very little difference should be seen between the coverage of human chromosome 20 by mouse chromosome 2 and by the whole mouse genome. The results are summarized in Table1 . The overall conclusion is that while SLAGAN may not be quite as sensitive as BLASTZ on the whole genome scale, it is able to align a larger percentage of the orthologous regions of human chromosome 20 and has higher specificity as shown by the lower percentage of non-orthologous alignments.... ..."

"... In PAGE 7: ...3 Results In the development of our testing procedures, we compared the performance of sev- eral previously published matrices and corresponding gap penalties optimized in reference 15 as applied to the 21 largest families of disorder ( Table1 ). The matrices were ranked by their cumulative scores (see 2.... In PAGE 8: ...and both original and modified all-positive scoring matrices. We also tried gap pen- alties 12/2 exploited in reference 24 but they exhibited poor performance and are excluded from Table1 , except for the OPTIMA matrix for which they are optimal. Overall, the results of Table 1 suggest that, of the previously published scoring matrices, the matrix of Gonnet et al.... In PAGE 8: ... We also tried gap pen- alties 12/2 exploited in reference 24 but they exhibited poor performance and are excluded from Table 1, except for the OPTIMA matrix for which they are optimal. Overall, the results of Table1 suggest that, of the previously published scoring matrices, the matrix of Gonnet et al.22 performed the best on our disordered protein families with less than 50% sequence identity.... In PAGE 8: ...alues were tested in steps of 0.25 in the interval from 1 to 2). This new, disorder- specific scoring matrix (Fig. 3) differs significantly from all of the other scoring matrices in Table1 . For example, DISORDER differs in 100 out of 210 positions (47.... In PAGE 9: ...he distribution of scores for different HMMs were compared (Fig. 4). Shaded bars represent the number of test proteins for which the DISORDER matrix obtained higher scores when aligned to the appropriate HMM, while white bars represent the same number for the BLOSUM62 matrix. These comparisons are plotted as a function of average sequence identity (quantized into 10 bins) as defined above, in the de- scription of Table1 , but without any threshold. 10 20 30 40 50 60 70 80 90 100 0 20 40 60 80 100 120 Number of test sequences with higher scores Sequence identity (%) Figure 4.... ..."

"... In PAGE 7: ...3 Results In the development of our testing procedures, we compared the performance of sev- eral previously published matrices and corresponding gap penalties optimized in reference 15 as applied to the 21 largest families of disorder ( Table1 ). The matrices were ranked by their cumulative scores (see 2.... In PAGE 8: ...and both original and modified all-positive scoring matrices. We also tried gap pen- alties 12/2 exploited in reference 24 but they exhibited poor performance and are excluded from Table1 , except for the OPTIMA matrix for which they are optimal. Overall, the results of Table 1 suggest that, of the previously published scoring matrices, the matrix of Gonnet et al.... In PAGE 8: ... We also tried gap pen- alties 12/2 exploited in reference 24 but they exhibited poor performance and are excluded from Table 1, except for the OPTIMA matrix for which they are optimal. Overall, the results of Table1 suggest that, of the previously published scoring matrices, the matrix of Gonnet et al.22 performed the best on our disordered protein families with less than 50% sequence identity.... In PAGE 8: ...alues were tested in steps of 0.25 in the interval from 1 to 2). This new, disorder- specific scoring matrix (Fig. 3) differs significantly from all of the other scoring matrices in Table1 . For example, DISORDER differs in 100 out of 210 positions (47.... In PAGE 9: ...he distribution of scores for different HMMs were compared (Fig. 4). Shaded bars represent the number of test proteins for which the DISORDER matrix obtained higher scores when aligned to the appropriate HMM, while white bars represent the same number for the BLOSUM62 matrix. These comparisons are plotted as a function of average sequence identity (quantized into 10 bins) as defined above, in the de- scription of Table1 , but without any threshold. 10 20 30 40 50 60 70 80 90 100 0 20 40 60 80 100 120 Number of test sequences with higher scores Sequence identity (%) Figure 4.... ..."

"... In PAGE 8: ... Indeed, in situations where the seed alignment is constructed from dif- ferent subfamilies, sequence motifs common to a subset of neighbors but not present in the test-set domain can only be aligned correctly using multiple alignment tools. To examine the effect of gap content on PSSM recogni- tion sensitivity we compare the recognition sensitivity of PSSMs derived from alignments with approximately the same fraction of gaps, as shown in Table2 . Only alignments with average sequence identity below 30% are included.... In PAGE 8: ... Only alignments with average sequence identity below 30% are included. As can be seen from Table2 , seed alignments with more gaps produce less sensitive PSSMs for all alignment algorithms, presumably because the sequences in these alignments are among the most diverse. Interestingly, for seed alignments containing equal fractions of gaps, the local alignment methods perform as well as the global alignment methods, and the local-structure (VAST) method gives the most sen- sitive PSSMs.... ..."

"... In PAGE 4: ... No statistically signi#0Ccantpowers were seen in any of the 48 acceleration trials. The resulting upper limit is #18 40 mCrab pulsar #0Dux units #28see Table2 #29. For comparison, the sensitivityofanepoch-folding search with known frequency... ..."