The Effects of Missing Taxa and Phylogenetic Imputation on Spatial Phylogenetic Studies
Luis Francisco Henao‐Diaz, Shawn W. Laffan, Brent D. MishlerABSTRACT
Aim
Spatial phylogenetic tools offer an evolutionary and geographical picture of biodiversity by characterising its multidimensional facets. Although phylogenetic trees are becoming increasingly comprehensive, they still represent only a fraction of the known terminal taxa in any biota. As an attempt to improve sample coverage, missing taxa are frequently imputed into a backbone phylogeny using their taxonomic classification. Here, we evaluated the effects of both incomplete taxon sampling and phylogenetic imputation on the quantitative, spatial patterns of phylodiversity.
Location
Japan.
Time Period
Present day.
Major Taxa Studied
Ferns (Polypodiopsida).
Methods
Using a dataset that is nearly complete for known terminal taxa, we conducted sensitivity analyses for a set of phylodiversity and endemicity analyses. We analysed trees using two approaches: (1) an increasing proportion of tips pruned under different selection scenarios, and (2) those tips phylogenetically imputed back into the tree using taxonomic classification. Tips were pruned at random and also using two methods that mimicked: (1) biased sampling based on species range sizes and (2) biased sampling among major clades on the tree. We assessed predictive performance and bias across spatial phylogenetic metrics and endemicity categorizations.
Results
Across all pruning methods, phylogenies with considerably reduced taxon sampling (particularly with biased sampling) exhibited lower predictive performance in endemicity categorization than imputed trees. However, unbiased incomplete trees had lower mean error for observed values of all phylodiversity metrics, as compared to imputed trees and SES metrics (which were also highly volatile). Estimation biases are lowest in trees with taxa imputed at random below‐crown nodes.
Main Conclusions
Our work suggests exercising caution when conducting spatial phylogenetic analyses in cases of limited taxon sampling, particularly when such sampling is spatially biased. Careful taxonomic imputation at random below a crown node is an acceptable solution, especially if sampling is spatially biased against geographically range‐restricted taxa. Tip imputations must be done under a probabilistic framework, using multiple imputations, and regarded as temporary hypotheses instead of a replacement for empirical data.