Fill in Missing Species Traits Using a Phylogenetic Tree

Imputes missing species trait data for comparative analyses by combining three sources of information: phylogenetic similarity (closely related species share similar traits), cross-trait correlations (observed traits inform missing ones), and optional environmental covariates (climate, habitat, geography). Handles continuous measurements, counts, binary variables, ordered categories, unordered categories, bounded proportions, zero-inflated counts, and compositional multi-proportion data in a single call. The default method is a phylogenetic baseline calibrated on held-out data; an optional graph neural network correction, which is needed to use covariates, is blended in through a per-trait gate that retains a baseline fallback. Provides conformal prediction intervals for continuous, count, ordinal, and proportion traits and an experimental analysis-aware multiple-imputation workflow for one missing continuous covariate in Gaussian linear, binomial-logit, and Gaussian random-intercept models, with Rubin pooling limited to fixed effects. Stochastic graph-network and posterior-tree completions are prediction diagnostics rather than validated inferential imputations. Bundled datasets include 300-species and 9,993-species bird-trait subsets with matching example phylogenetic trees. Rubin (1987, ISBN:978-0-471-08705-2); Vovk et al. (2005, ISBN:978-0-387-25061-8); Nakagawa and de Villemereuil (2019) .

pigauto: Phylogenetic Imputation via Graph AUTO-encoders

Warning: pigauto is experimental; use at your own risk. It needs further validation. Point estimates are the supported claim. Prediction intervals are nominal held-out diagnostics, not package-wide certification; covariance routes have focused-test evidence only.

pigauto fills missing species trait values using a phylogenetic tree, other traits, and optional covariates. It supports continuous, count, binary, categorical, ordinal, proportion, zero-inflated-count, and compositional traits.

Installation

The workflow below uses functions and defaults from pigauto 0.11.0. Check packageVersion("pigauto") after installation: the CRAN release can lag behind the GitHub version described on this site.

For the released version:

install.packages("pigauto")
packageVersion("pigauto")

If CRAN has an earlier version, install the GitHub version to use this workflow:

# install.packages("pak") # once, if needed
pak::pak("itchyshin/pigauto")
# torch::install_torch() # once, if you opt into gnn = TRUE

Start here

Put a CSV trait table (with a unique species column) and a Newick or NEXUS tree beside your R script. Then start with this short workflow:

library(pigauto)

traits <- read_traits("traits.csv")
tree <- read_tree("tree.nwk")
check_pigauto(traits, tree)
result <- impute(traits, tree)
completed <- completed_data(result)
pigauto_report(result)

BirdTree distributes its tree files in Newick format. Download one from BirdTree and read it with read_tree(); read_tree() also accepts NEXUS files from other sources. Cite Jetz et al. (2012) for the tree, and cite BirdTree.org when you use its website. pigauto uses the tree you provide and does not download BirdTree data automatically. The bundled tree300 is a teaching example; it is not a consensus or recommended analysis tree.

impute() now uses the phylogenetic baseline by default (gnn = FALSE), estimates Pagel's lambda for eligible traits, and leaves the mean safety floor and phylogenetic-signal gate off. To train the optional graph neural network, install the torch runtime once and call impute(traits, tree, gnn = TRUE). The former GNN-on behaviour is an explicit choice in 0.11.0.

check_pigauto() and completed_data() were added in 0.11.0. If your installed version is 0.10.0, call impute(traits, tree) directly; that version does not include check_pigauto().

check_pigauto() runs before fitting: it reports input errors, species/tree matching, trait declarations, and estimated size. Its default gnn = FALSE matches impute() and does not probe torch. Pass gnn = TRUE when planning a GNN fit to include runtime availability. Resolve an "error" status before fitting; a fully observed target table does not need imputation, so use cross_validate() instead.

The output roles are deliberately separate:

• completed is the input-shaped trait data with observed values retained and modeled missing cells filled.
• result$prediction contains all-cell diagnostic predictions, not a second completed dataset.
• result$prediction$se is type-dependent uncertainty; discrete values are not Gaussian standard errors. Conformal bounds are nominal held-out diagnostics.
• Split conformal intervals assume the calibration cells resemble the scored cells. Masking a whole clade breaks that assumption because calibration then comes from other clades.
• A pigauto_result does not itself authorize downstream inference. The supported pooling entries begin with multi_impute_analysis() in its documented narrow regime or, for continuous traits, multi_impute(draws_method = "posterior"). multi_impute()'s default, draws_method = "auto", picks "posterior" whenever the data allow it and otherwise says why it fell back to non-poolable conformal draws.

Tiny installed example inputs are available at system.file("extdata", "novice_traits.csv", package = "pigauto") and system.file("extdata", "novice_tree.nwk", package = "pigauto"). The matching installed six-expression script is system.file("examples", "novice-workflow.R", package = "pigauto").

Copy-ready data recipes

Ordered ecological states

Use an ordered factor when states have a known order.

traits$threat_status <- ordered(
  traits$threat_status,
  levels = c("LC", "NT", "VU", "EN", "CR")
)
result <- impute(traits, tree)

Integer-valued continuous measurements

An integer column is otherwise read as a count. Declare a measurement stored as whole numbers explicitly.

result <- impute(
  traits, tree,
  trait_types = c(body_length_mm = "continuous")
)

Proportions and zero-inflated counts

Declare proportions only when their numeric 0–1 scale is semantically a proportion. Declare zi_count only when zero is a separate process from the positive count magnitude.

result <- impute(
  traits, tree,
  trait_types = c(habitat_cover = "proportion", parasite_load = "zi_count")
)

Compositions

Each declared composition is one trait: every row must have all components or all components missing, and observed rows must sum to one.

result <- impute(
  traits, tree,
  multi_proportion_groups = list(diet = c("diet_insect", "diet_fruit", "diet_seed"))
)

Repeated observations

read_traits() requires unique species names. For repeated observations, read the CSV directly and supply species_col.

traits_obs <- read.csv("traits_repeated.csv", stringsAsFactors = FALSE)
check_pigauto(traits_obs, tree, species_col = "species")
result <- impute(traits_obs, tree, species_col = "species")

Reconciling species names

Inspect the structured species report before changing data:

check <- check_pigauto(traits, tree)
check$species$matched
check$species$data_only
check$species$tree_only

data_only rows remain in completed but are not modeled or filled. tree_only tips are internal all-missing rows used only while fitting.

Defaults, and when to change them

impute(traits, tree) with its defaults is the starting point. The table says which route to take for each need; reach for another route only for the need in the left-hand column.

For other data, draws_method = "auto" falls back to "conformal" draws and says so. Those draws, like "mc_dropout", describe the spread of plausible values but are not proper multiple imputations: in a simulation of a phylogenetic regression they biased the pooled slope towards zero, and with_imputations() refuses them. Bayesian downstream fits (brms, MCMCglmm) are combined by stacking posterior draws, for example with brms::brm_multiple(), not with pool_mi(). Worked examples for nlme, lme4, glmmTMB, drmTMB, gllvmTMB and brms are in the multiple imputation article. The ability to fit and extract coefficients from a model class does not by itself validate the imputation model for that analysis; check the article's conditions before interpreting pooled estimates.

Advanced controls

The default journey above is the supported starting point. These controls are for an explicit technical workflow and are kept separate from it.

# Solver, EM, and refinement controls
result <- impute(
  traits, tree,
  joint_solver = "rphylopars",
  em_iterations = 2L,
  joint_refine_iter = 1L,
  conformal_split_val = TRUE
)

# Prediction route: the default "auto" chooses per trait between the
# cross-trait "exact" conditional and "per_column"; force one with
result <- impute(traits, tree, predict_method = "exact")

See ?impute, ?fit_baseline, and the articles for all calibration, refinement, and prediction controls.

More help

• Getting started
• Mixed-type traits
• Multiple imputation and downstream models
• Common pitfalls
• Tree prediction sensitivity

Citation

Nakagawa S (2026). pigauto: Phylogenetic Imputation via Graph Autoencoder.