I am aligning merged amplicon reads (around 280-300 bp) to an initial alignment of a merged read and a reference gene for several samples using MAFFT and the --addfragments option. It works just fine for the great majority of samples (20/24), but for some I get the very uninformative error 'hat2 is wrong', resulting in an empty output file. I am aware of Mafft Output Distance Matrix and https://www.biostars.org/p/184512/, but these do not really provide information about the error. The command I run is
mafft --addfragments read_file --keeplength --reorder --thread 40 alignment_in > alignment_out
and the error:
Loading 'hat2n' (aligned sequences - new sequences) ... 109626 != 10962
hat2 is wrong
Since hat2 seems to encode a distance matrix and the sequence numbers match except for the last number, is it possible that the matrix is somehow truncated? The strange thing is that I do not get the error with reads from the same sample when I map them to another reference gene. Other samples however work with that reference gene but not with the other.
Does somebody have a clue what is going on here?