Evolutionary shift detection with ensemble variable selection

Abstract

Abrupt environmental changes can lead to evolutionary shifts in trait evolution. Identifying these shifts is an important step in understanding the evolutionary history of phenotypes. The detection performances of different methods are influenced by many factors, including different numbers of shifts, shift sizes, where a shift occurs on a tree, and the types of phylogenetic structure. Furthermore, the model assumptions are oversimplified, so are likely to be violated in real data, which could cause the methods to fail. We perform simulations to assess the effect of these factors on the performance of shift detection methods. To make the comparisons more complete, we also propose an ensemble variable selection method (R package ELPASO) and compare it with existing methods (R packages [Formula: see text]1ou and PhylogeneticEM). The performances of methods are highly dependent on the selection criterion. [Formula: see text]1ou+pBIC is usually the most conservative method and it performs well when signal sizes are large. [Formula: see text]1ou+BIC is the least conservative method and it performs well when signal sizes are small. The ensemble method provides more balanced choices between those two methods. Moreover, the performances of all methods are heavily impacted by measurement error, tree reconstruction error and shifts in variance.

Keywords: ELPASO; Ensemble method; Evolutionary shift detection; LASSO; Ornstein-Uhlenbeck model; Phylogenetic comparative methods; Trait evolution.

Fig. 1

The model structure of ensemble method for shift detection

Fig. 2

Tree used in simulations to compare the precision and recall of different methods. The shifts positions are indicated by asterisks. Different colours indicate different optimal values for the trait

Fig. 3

True positive numbers versus False positive numbers with 7 shifts

Fig. 4

The mean log likelihood on 1000 test datasets

Fig. 5

ARI with 7 true shifts

Fig. 6

Shifts in different positions of tree

Fig. 7

True positive versus false positive with different shift positions

Fig. 8

Four different types of tree

Fig. 9

True positive versus false positive with different types of trees

Fig. 10

Average test log likelihood with parameters estimated from training data without measurement error using the shifts selected from training data with measurement error

Fig. 11

Regenerated tree with different beta

Fig. 12

Number of non-converging cases when analyzing on a misspecified tree

Fig. 13

True positive versus false positive with applying methods on misspecified trees (larger triangles show larger difference with the original tree)

Fig. 14

Average test log likelihood with applying methods on misspecified trees

Fig. 15

Average test log likelihood with changing estimated alpha

Fig. 16

Diffusion variance parameter changes on branch 195

Fig. 17

Diffusion variance parameter changes on branch 195