Version 4 release notes

User-facing changes in the version 4 series, newest first. See NEWS.md for the full change history.

Version 4.3 Current release

New capabilities and improvements

  • Locate positive selection, even when the alignment is uncertain. Earlier versions could detect evidence of positive selection in a gene; 4.3 can also identify the codons involved, averaging over alternative alignments.
  • Visualize evolutionary properties on alignments. Color shading displays estimated rates, dN/dS, and positive-selection probabilities for individual letters or codons. Under supported heterotachy models, it can reveal rate differences between letters in the same column.
  • A fuller range of codon analyses. Run single-ω analyses, site tests, branch tests, and branch-site tests using Bayesian inference. New BranchModel and BranchModel_test models estimate and test differences in dN/dS between branch groups. The guide provides example commands and explains how to interpret the results.
  • Less Monte Carlo noise in positive-selection support estimates through Rao–Blackwellization.
  • Gamma-distributed rate variation: quadrature as an alternative to equal-probability bins.
  • Rename Rates.* → ASRV.* (old names remain as deprecated aliases).

Corrections

  • Correct a proposal-probability calculation in SPR tree moves.
  • Fix likelihood rescaling for fixed-alignment analyses.
  • Fix the multi-nucleotide mutation model for codons.
  • Correct the FMutSel prior to allow larger fitness differences.
  • Prevent crashes when proposals produce infinite or undefined values.
  • Give more helpful error messages for invalid Newick trees.
  • Preserve underscores in Newick tree labels instead of converting them to spaces.

Breaking changes

  • Capitalize model and distribution names: gtr → GTR, normal → Normal.
  • Rename bp-analyze → bpy-summarize.
  • Rename bali-subsample → bpy-subsample.
  • Unconstrained single-ω codon models now use a LogNormal(0,1) default prior, allowing ω above as well as below one.

Version 4.2

New capabilities and improvements

  • Faster likelihood calculations for models with many states, including a fourfold speedup for codon models.
  • Faster startup and model execution.
  • Amino-acid profile mixture models C10, C20, …, C60.
  • Amino-acid models LG4M and LG4X.
  • Log departures from reversibility in terms of flux and nonreversibility.
  • Inspect indel size, location, and sequence with joint-indels.

Corrections

  • Fix non-reversible models combined with Markov-modulated models.
  • Improve amino-acid alphabet detection when sequences contain many A, T, G, C, or N characters.
  • Fix model parsing in Bioconda ARM builds.

Version 4.1

New capabilities and improvements

  • Compare multiple consensus alignments in the analysis summary.
  • Condition on variable sites with --smodel='MODEL|variable'.
  • Faster SPR tree moves on large trees.

Corrections

  • Fix strict consensus-alignment construction.
  • Fix character ranges selecting first, second, or third codon positions.
  • Fix proposed alignments in SPR moves.
  • Fix default alphabet and substitution-model selection for binary data.
  • Fix summary generation for the Numeric(2) alphabet.

Version 4.0: additions since version 3

These highlights include capabilities introduced during the 4.0 beta series.

New capabilities and improvements

  • A probabilistic programming language for specifying custom models.
  • BUSTED and BUSTED-S codon models.
  • Multi-nucleotide mutation codon models.
  • Support for non-standard genetic codes.
  • Non-reversible and non-equilibrium substitution models.
  • Branch-specific insertion-deletion rates.

Corrections

  • Fix excessive memory usage on large trees in versions 3.5–3.6.