Modisco-lite

==============

report.py
---------

- Removes the use of the Arial Rounded font, hopefully reducing the number
of warnings produced.

==============

report.py
---------

- Removed the need to supply both a MEME file with motifs and a directory
with extracted PFMs. The PFMs will be read directly from the MEME file now.


modisco
-------

- Removed the need to supply both a MEME file with motifs and a directory
with extracted PFMs. The PFMs will be read directly from the MEME file now.

==============

report.py
---------

- Fixed a minor error where results would not display properly if fewer
than the specified number of TOMTOM hits were found for a motif.

==============

report.py
---------

- Fixed a display issue where the patterns were being displayed in the
wrong order.

==============

Highlights
----------

- A minor bug and a minor convenience issue have been resolved.


aggregator.py
-------------

- A minor bug was fixed where, when collapsing patterns in the final step,
some patterns would have no seqlets after the filtering step and the code
would crash. Now, if a pattern has no seqlets after the filtering, it is
ignored.


report.py
---------

- The patterns are now displayed in numeric order 1-10, followed by 11-20,
etc., as opposed to by alphabetical order 1, 10-19, 2, 20-29, etc.

==============

Highlights
----------

- This is the first complete release that includes breaking changes compared
with the original code. They are detailed below.


affinitymat.py
--------------

- When calculating the coarse-grained resolution, an argsort is done to find
the nearest neighbors of each seqlet. This is done in numpy using a sorting
algorithm that is not stable. This new version uses "mergesort" instead of
the default "quicksort", as it is a stable algorithm. This change allows
us to use numpy within numba, assigning more work to each thread, and
ultimately speeding the code up ~2-3x.


core.py
-------

- AggregatedSeqlet was renamed SeqletSet.


gapped_kmer.py
--------------

- A bug in the original codebase where the nucleotide identity was done
on the argmax of the contribution score, rather than the one-hot encoded
sequence, caused incorrect nucleotide calls for positions that had a
negative contribution. This effects only the gapped k-mer representation
used to calculate coarse-grained representations, not the found patterns.


io.py
-----

- This code was added to handle saving the outputs of the TFMoDISco function
to disk and the conversion of the old outputs into the new format.


tfmodisco.py
------------

- The output of TFMoDISco is now a tuple of two lists, where the first
list contains the positive patterns and the second list contains the
negative patterns. Each pattern is represented as a `SeqletSet` object.