NEWS

---

evobiR 2.2.1

New features

• AncCond() accepts missing data: x and y may each be observed on any subset of tips, including disjoint subsets (previously any missing value was an error). Tips without y are marginalised in the Mk likelihood and the stochastic maps, and null simulations are masked to the observed missingness pattern. Nodes without a direct continuous estimate receive the Brownian-motion conditional expectation given the x-bearing tips (branch-length interpolation along the pruned-tree edge; attachment-point value for dataless subtrees). The result gains an n.obs element. Note for prune = TRUE: nodes in state-1 subtrees now take this conditional expectation instead of the previous fallback, which used a fastAnc fit to all tips including the state-1 lineages that prune is meant to exclude.

Bug fixes

• make.simmap2() now indexes the transition matrix by character state name everywhere instead of treating state labels as numeric row indices. Traits coded 0/1 previously failed with "no positive probabilities"; states with labels like chromosome numbers ("5", "12", ...) could silently index the wrong rows. User-supplied Q matrices without dimnames are now given state names automatically.
• make.simmap2() now honors rejmax, rejint, and monitor for multiPhylo input (it previously fell back to phytools::make.simmap, silently dropping these arguments).
• countTrees() now works with single-tree collection or reference files (previously indexed tree components on phylo inputs).
• SampleTrees() burn-in no longer retains the last burn-in tree, no longer produces fractional/zero indices, validates burnin and final.number, and writes correctly named output files (previously "prefix .nwk" with a space).
• scaleTreeRates() returns class c("phyloscaled", "phylo") so that plot() dispatches to plot.phyloscaled() (scalars were previously silently ignored in plots); tie-breaking in the final-likelihood selection now uses which.max().
• GetTipRates() stops with an informative error naming missing tips when tip labels cannot be matched (previously crashed with if (NA) or proceeded on misordered data after a printed note).
• ResSel() now selects the same number of individuals in both tails (the low tail previously selected one fewer); plotting no longer assumes the two trait columns are adjacent and ascending; an unsupported model now errors informatively.
• WinCalcD() returns properly typed (numeric) columns instead of all-character ones and no longer drops the final window.
• CalcD()/CalcPopD() jackknife now drops exactly block.size contiguous sites (previously block.size + 1) at whole-number positions (previously fractional).
• fix.simmap() correctly resolves start states for failed tip branches whose parent is the root, no longer leaks the working table across simulations, and the shared-leading-edge test no longer always fires (self-match).
• SuperMatrix() ignores its own previous output files on re-runs, treats input as a literal string rather than a regular expression, and errors when no files match.
• getNe() errors informatively on an unrecognized locus value (previously "object 'ne' not found").

Performance improvements

• CalcD()/WinCalcD(): the per-site biallelic test is now fully vectorized (~20x faster; the gain multiplies under bootstrap replication).
• scaleTreeRates(): matrix exponentials are memoized by scaled edge length once Q is fixed, removing >95% of expm() calls.
• AncCond(): per-edge transition matrices are precomputed once and reused across all null simulations.
• CalcPopD(): population row indices are computed once per dataset rather than roughly ten times per site.

Documentation and metadata

• AncCond() documentation now matches the code defaults (n.sims = 1000, Q = NULL).
• Non-ASCII characters in R sources replaced with ASCII equivalents.
• rmarkdown added to Suggests (vignette builder requirement); unused imports removed from NAMESPACE; stats::rbinom imported; print.anccond registered as an S3 method; Mode() exported.
• Package Version bumped to 2.2.1.