smoothxg 0.6.4 - Pasticcione

Buildable Source Tarball: smoothxg-v0.6.4.tar.gz

This introduces minor fixes towards the conda build.

Smoothxg 0.6.3 - Generico

Buildable Source Tarball: smoothxg-v0.6.3.tar.gz

This fixes a compilation setting that was preventing odgi dependency to be built as requested with CMAKE_BUILD_TYPE.

Smoothxg 0.6.2 - Magrissimo

Buildable Source Tarball: smoothxg-v0.6.2.tar.gz

This release includes:

• smoothxg can display its current version;
• interface improvement, by grouping parameters by function;
• usage of an improved (cooled) version of the PG-SGD algorithm for graph sorting;
• updated odgi (faster and smarter compilation) and args dependencies;
• adaptive partial order alignment (POA) penalties with respect to the mash-estimated pairwise identity of the sequences in the block.

Smoothxg 0.6.1 - Magro

Buildable Source Tarball: smoothxg-v0.6.1.tar.gz

This release includes:

• block padding can be disabled;
• block padding is done with respect to the average sequence length of the block;
• do not pad blocks that are very deep (the threshold can be given in input) to avoid huge computational overheads in complex regions;
• disabled abPOA seeding: this improves accuracy and avoids alignment dropouts in low complexity regions;
• fixed how blocks are split with respect to the block depth.

unciampato

Buildable Source Tarball: smoothxg-v0.6.0.tar.gz

We have been tripped up by the complexity of evaluating the full POA problem for long sequences. Thanks to @yangao07's work on seeded POA alignment, we can reduce this requirement with little loss of alignment quality. Maintaining accuracy requires setting abPOA's min_w parameter about 60-fold higher than default (~3kb). The memory and runtime requirements become much lower in general, which is essential to scaling pggb to large problems.

This version also introduces "handy" human-readable parameters, e.g. 5k or 20m rather than 5000 or 20000000.

Smoothxg 0.5.2 - Imbottitura

Buildable Source Tarball: smoothxg-v0.5.2.tar.gz

This release includes:

• each block is padded on the left and on the right before the POA: this reduces the noise at the blocks' boundaries, and helps to stabilize the results;
• abPOA can be used also in local mode;
• the number of threads to use during the PO alignments can be specified, to limit the memory usage;
• a little compiling fix.

Smoothxg 0.5.1 - Scaling up

Buildable Source Tarball: smoothxg-v0.5.1.tar.gz

This point release adds an option to set a lower number of threads during POA. This can be important to make sure we don't get stuck in big pggb jobs.

stabilizing the smooth

There are certainly many improvements that we can make to the general approach of normalizing and locally compacting pangenome graphs. But, what we have here is a stable, scalable method. This checkpoints a lot of work that's happened since v0.4.

smoothxg 0.4 - Super

Buildable Source Tarball: smoothxg-v0.4.tar.gz

This introduces a massive cleanup of memory problems and timings issues.

The abPOA integration has been refactored to reduce memory consumption and time. In addition, the 'PO graph to odgi graph' conversion has been updated to avoid high memory consumption for bigger graphs with newer versions of odgi.

Several steps have been simplified and cleaned, including the consensus graph generation.

Furthermore, the block split has been strongly improved. Now very depth blocks can be split quickly, applying the edit-based sequence clustering. For particularly deep blocks, a new mash-based sequence clustering is applied as last resort. It is very fast, at the price of having resultings blocks with some noise as the mash-based identity estimations tend to over-estimate the real sequence identities.

The problems were tackled using AddressSanitizer and LeakSanitizer.

super smoothxg pre-release

An update to odgi greatly improves performance.

Many updates to smoothxg have reduced its memory footprint and runtime.