LAST and repeats
================

Biomolecule sequences are full of repeats, which have a massive effect
on sequence comparison.  There are two different things called
"repeats":

* Interspersed repeats, such as LINEs, Alus, transposons.

* Simple sequence / low-complexity sequence / short-period tandem
  repeats, such as atatatatatatatatat.

They cause two different problems:

* Simple tracts cause alignments of sequences that are not homologous
  (not descended from a common ancestor), because similar simple
  sequences can evolve independently.

* Both kinds of repeat can make sequence comparison very slow, and
  produce an overwhelming number of alignments.

LAST options
------------

LAST has built-in recognition of simple sequences, but not (yet)
interspersed repeats.  So you may wish to annotate interspersed
repeats with a tool such as RepeatMasker or WindowMasker.  You can
often obtain pre-masked genomes, with repeats indicated by lowercase.

LAST's built-in simple-sequence masking uses tantan_, which prevents
non-homologous alignments more reliably than previous methods.

To mask all kinds of repeat, use lastdb -c -R11:

  This keeps lowercase in the input sequences, additionally lowercases
  simple tracts found by tantan, and masks lowercase during alignment.

To mask simple tracts only, use lastdb -c -R01:

  This does not keep lowercase in the input sequences, lowercases
  simple tracts found by tantan, and masks lowercase during alignment.

If no -R option is specified, it defaults to -R01 (since LAST version
1205).  In LAST, "mask" means that lowercase is excluded when finding
similar sequences, but included when making the final alignments.

Orthology search and post-masking
---------------------------------

Sometimes we wish to find orthologs but not paralogs: e.g. if we
compare two whole genomes, or align DNA reads to a genome.  Masking
can cause false positives, if an ortholog is masked but a paralog is
not.

We'd like to avoid this problem, but also avoid non-homologous
alignments of simple sequences.  This can be done by post-masking:

1. Align using lastdb -R01 (the default): this does not keep lowercase
   in the input sequences, lowercases simple tracts found by tantan,
   but does not mask it (i.e. case has no effect).

2. Predict orthologs, e.g. using `last-split <doc/last-split.rst>`_.

3. Discard mostly-lowercase alignments, using `last-postmask
   <doc/last-postmask.rst>`_.

.. _tantan: https://gitlab.com/mcfrith/tantan