Clustering large genomes with psi-cd-hit giving a different similarity value to blast

Question

[ang.mccann]

Hi, I am trying to cluster bacterial genomes with psi-cd-hit but I have some results I can't understand. I run the command as follows

./psi-cd-hit.pl -i Genomes.fasta -o Genomes.fasta.k.out. -c 0.8 -G 1 -g 1 -prog blastn -circle 1

However, when I check the clustered output file the following is a snippet from my results

>Cluster 3
0       991702aa, >NC_018077.1... *
1       991702aa, >CP002058.1.. at 0.0/991702aa/100%
2       977322aa, >NZ_CP007589.1.. at 0.0/672875:123109:1407:1041:348:978:346:249:90:52aa/81.82%

So sequences 0 and 1 are identical, which is correct, and sequences 0 and 2 are 82% similar according to psi-cd-hit. However, based on blast they are about 98% identical. Here is a snippet of the blast output

blastn -query a.fasta  -subject b.fasta  -outfmt 6

 NC_018077.1    NZ_CP007589.1   99.99   672875  31  34  181458  854328  181375  854215  0   1.24E+06
 NC_018077.1    NZ_CP007589.1   99.99   181393  4   8   1   181393  1   181385  0   3.35E+05
 NC_018077.1    NZ_CP007589.1   99.98   123109  11  6   868600  991702  854216  977322  0   2.27E+05
 NC_018077.1    NZ_CP007589.1   99.69   4905    10  5   708933  713835  715477  720378  0   8970
 NC_018077.1    NZ_CP007589.1   99.65   4906    10  6   715583  720483  708827  713730  0   8959

So, although I didn't see it in the manual I believe the numbers after each genome in the psi-cd-hit cluster output are the segments of sites that pass the -c threshold i.e.

>NZ_CP007589.1.. at 0.0/672875:123109:1407:1041:348:978:346:249:90:52aa/81.82%`

means a block of 672875 passed, followed by a block of 123109 sites etc and if I sum these sites and divide by the genomes length I get 82%. When I look to the blast output it also shows a block of length 672875 at 99% identity and a block of 123109 but the block of length 181393 is not present in the psi-cd-hit output and may be accounting for the lower score (82% by psi-cd-hit). Am I missing something. I have seen a similar question on this Remove redundancy from GenBank plasmid database using cd-hit-est but this was using cd-hit-est and I am using the -circle option with psi-cd-hit Can anyone help Thanks

[h.mon]

Hi, I am trying to cluster bacterial genomes with psi-cd-hit but I have some results I can't understand.

Are you sure about what you are doing? From the documentation:

• psi-cd-hit Cluster proteins at <40% cutoff

Ok, sorry, just read the docs more carefully, and it is indeed an envisaged use case.