According to the documentation provided online, MAFFT uses L-INS-i - Iterative refinement methods using WSP and consistency scores when --localpair option is used.

PS: Also check out the parallelized version of MAFFT if you are trying to align a large number of sequences.

This question is somewhat old, but I've run into the exact same problem! Here are all of my attempts to correctly specify the --MSA_Param argument in order to set the "maxiterate" parameter to 1000 and enable the "localpair" option when using GUIDANCE2 (V2.02) on a command line:

--MSA_Param \-\-maxiterate 1000 \-\-localpair

--MSA_Param \--maxiterate 1000 \--localpair

--MSA_Param --maxiterate 1000 --localpair

--MSA_Param \-maxiterate 1000 \-localpair

\-maxiterate 1000 \-localpair

\-\-maxiterate 1000 \-\-localpair

--maxiterate 1000 --localpair

So far, none of them have worked. I've emailed the creators of GUIDANCE2 for assistance and I'll let you know if they respond.

EDIT: I think I figured it out! Specifying my desired GUIDANCE2 options in this way exactly didn't return an error

--MSA_Param \\--localpair \\--maxiterate 1000

In my case, I had to specify the maxiterate parameter after the localpair parameter. Otherwise, GUIDANCE would interpret the characters "localpair" as input to the maxiterate parameter. An return with a "non-integer argument" error.

EDIT #2: The creator of GUIDANCE2 got back to me and said the following:

Please wrap the --MSA_Param arguments also with a ' or " around all arguments (separated by whitespace) to designate they all should be passed to the MSA algorithm as a single argument. For example, in your command this would apply as follows:

--MSA_Param '\-\-maxiterate 1000 \-\-localpair'

This answer is a couple years too late, but I hope it's helpful to someone :)