Multiple alignment program for amino acid or nucleotide sequences

About

MAFFT is a multiple sequence alignment program for unix-like operating systems.  It offers a range of multiple alignment methods, L-INS-i (accurate; recommended for <200 sequences), FFT-NS-2 (fast; recommended for >2,000 sequences), etc. 

Merits

1. (Accuracy)
   L-INS-i is one of the most accurate multiple sequence alignment methods currently available.  L-INS-i is in particular suitable to align 10-100 protein sequences, because of an objective function combining the WSP and consistency scores. 

   Protein benchmarks are in:

   • Sievers et al. 2011
   • Simmons et al. 2008
   • Golubchik et al. 2007
   • Ahola et al. 2006
   • Nuin et al. 2006
   • Pei et al. 2006
   • Roshan & Liversay 2006
   • Armougom et al. 2006

   RNA benchmark:

   • Wilm et al. 2006

   DNA benchmark:

   • Carroll et al. 2007

   Phylogeny-based benchmark:

   • Dessimoz & Gil 2010
   • Letsch et al. 2010

2. (Scalability)
   FFT-NS-2 and other progressive methods can align many and/or long DNA/protein sequences, because of an FFT approximation and a linear-space DP algorithm. 
3. The scoring system was designed to allow large gaps.  Thus MAFFT is suitable for LSU rRNA and SSU rRNA alignments that sometimes have variable loop regions.  Staggered gaps (like the figure below) are also allowed.  This feature is remarkable with the --addfragments option.

   
   

Limitations

1. (Accuracy)
   Library extension is not performed unlike TCoffee and ProbCons-CONTRAlign, because we think at present that iterative refinement is more efficient than library extension. 
2. (Scalability)
   If two unrelated and long genomic DNA sequences are given, FFT-NS-2 tries to make a full-length alignment using rigorous DP and requires large CPU time.  For such a case, homology search tools such as FASTA and BLAST are more suitable. 
3. The order of alignable blocks or domains are assumed to be conserved for all input sequences.