Releases · pierrebarbera/epa-ng

Interface Related Changes

changed CLI library from cxxopts to CLI11
removed the --repeats option, now repeats are linked to whether or not masking is used (no masking = yes repeats)
removed the -O option to re-optimize the tree branch lengths and model parameters (this should be done upstream, then pass the parameters via --model)

Feature Changes

can work with amino acid data again
removed the experimental "pipeline" parallelization mode
added masking heuristic, enabled by default
slightly changed the bfast format to accomodate the masking information
changed default value of accumulated LWR candidate selection heuristic to 0.99999
added --model argument that takes either a raxml-ng style model descriptor specifying the exact model parameters to be used, or a RAxML_info file (raxml 8.x) from a -f e run
added pplacers baseball heuristic, for now with static default settings (--strike-box 3.0, --max-strikes 6, --max-pitches 40)
added explicit --no-heur flag that disables all preplacement heursitics (does not disable the new masking heuristic!)
added --no-pre-mask flag to disable masking heuristic
added --split function that takes a reference MSA file and one or more query aligment files, in phylip format, and outputs a fasta file containing only aligned query sequences. Usage: epa-ng --split ref_alignment query_alignments+. (this is a tailored function to prepare papara output for epa-ng) (issue #16)

Performance Changes

greatly increased parallel efficiency
added aforementioned masking heuristic that greatly increases overall performance
many smaller performance improvements

Bugfixes

fixed non-working --version option (issue #13)
fixed issues with RNA data (issue #14)
fixed a segfault that occured when input query sequence volume got too high (issue #9)
fixed a bug that allowed empty pqueries to sometimes be printed to the jplace

Code Changes you probably don't care about

updated genesis and libpll/pll-modules to newer commits
whole bunch of refactoring
re-write of the jplace output module, supports MPI I/O operations now to do massively parallel writes, asynchronously. Now scales indefinetly with input volume
separate, and reworked, mid-level placement functions for preplacement and thorough placement. (treating them differently made a big difference performance wise)
added a layer of separation to the readers to enable better automatic file format detection
switched MSA_Stream to use genesis instead of pll for reading
altered behavior of MSA_Stream slighly
greatly improved performance lwr calculation
decreased reliance on Work container
ressurected the Range and call_focused code :spooky: and used it to implement masking

Provide feedback