Positive-unlabeled host prediction via xplus (PLUS algorithm)

Predicts the hosts of a parasite from the phylogeny using positive-unlabeled (PU) learning, as a parallel alternative to the pairwise logistic model. Known hosts are treated as positives and every other species in the phylogeny as unlabeled (never as a true absence). Each species is described by phylogenetic features and the PLUS algorithm (xplus::xplus) iteratively re-labels the unlabeled species while fitting a penalized logistic model.

Usage

fit_phylopred_pu(
  interactions,
  phydist,
  parasite,
  n_eigen = 10,
  seed = NULL,
  ...
)

Arguments

interactions: A data.frame whose first two columns are the parasite and host species names (as in prepare_pair_data).
phydist: Square numeric matrix of phylogenetic distances with species names as row and column names.
parasite: Character scalar: the parasite whose hosts are predicted.
n_eigen: Number of PCoA axes used as features. Default 10.
seed: Optional random seed passed to xplus::xplus.
...: Further arguments passed to xplus::xplus (e.g. max_iter, learning_rate, qq).

Value

An object of class "phylopred_pu": a list with

prediction: data.frame with species, pred (predicted host suitability) and is_host.
fit: The underlying xplus object.
parasite: The parasite name.

The prediction columns can be passed directly to host_threshold_metric and phylo_partial_roc.

Details

The feature matrix contains, for every species in phydist:

the minimum and mean phylogenetic distance to the parasite's known hosts (excluding the species itself), and
the first n_eigen principal coordinates (PCoA) of the phylogenetic distance matrix, describing the global position of the species in the phylogeny (the analogue of environmental layers in an SDM).

Examples

pu <- fit_phylopred_pu(beetleTreeInteractions, phy_dist,
                     parasite = beetleTreeInteractions[[1]][1],
                     max_iter = 20, seed = 1)
host_threshold_metric(pu$prediction$pred, pu$prediction$is_host)
#> Phylopred threshold metric f_gamma (gamma = 1)
#> Optimal threshold: 0.6020  f_gamma: 39.1667
#> Evaluated at 190 thresholds.