Getting the best out of T-Coffee




  • T-Coffee
    1. t_coffee dataset.fasta
      Produces an alignment with an accuracy of 22.5%

    2. t_coffee dataset.fasta -method slow_pair
      Produces an alignment with an accuracy of 27%

    3. t_coffee dataset.fasta -method slow_pair slow_pair@EP@MATRIX@pam250mt
      Produces an accuracy of 25.6

    4. t_coffee dataset.fasta -aln ref_aln.aln -method slow_pair
      Produces an alignment with an accuracy of 27.1%. This lack of improvement over the 'fair' command line results from the wide disagreement between all the pairwise alignments. A single pair is not enough to compensate. As an alternative, you can overweight the library.

      • t_coffee -aln ref_aln.aln -lib_only -out_lib reflib -weight 1000
      • t_coffee dataset.fasta -lib reflib -method slow_pair
        This time we get an alignment 40% correct

    6. t_coffee -profile prf1 prf2 -method slow_pair
      has an accuracy of 32% (without cheating!)

    7. t_coffee -profile prf1 prf2 -aln ref_aln.aln
      has an accuracy of 100% (OK we cheat a 'little' here!)

      8. Note that in the last exemple, the profiles are threaded onto the reference alignment. This is the ideal way to force an alignment that you know is correct onto a bunch of homologous sequences.

    		•

    Questions should be sent to C.Notredame