Comparison of commonly used methods for combining multiple phylogenetic data sets

Transcription

1 Comparison of commonly used methods for combining multiple phylogenetic data sets Anne Kupczok, Heiko A. Schmidt and Arndt von Haeseler Center for Integrative Bioinformatics Vienna Max F. Perutz Laboratories June 12th, 2008

2 Motivation Multi-Locus Datasets Taxa Genes A B C D E F G H I J K L M N O P Q R S T a b c d e f g h i j k l m n o p q r s t data collection phylogeny reconstruction

3 Motivation Multi-Locus Datasets Taxa Genes A B C D E F G H I J K L M N O P Q R S T a b c d e f g h i j k l m n o p q r s t data collection phylogeny reconstruction

4 Motivation Multi-Locus Datasets Taxa Genes A B C D E F G H I J K L M N O P Q R S T a b c d e f g h i j k l m n o p q r s t Approaches: data collection phylogeny reconstruction early level medium level late level

5 Methods Early-level combination Early-level combination: Superalignment = Supermatrix or Total Evidence Combination by concatenating data sets: Any tree reconstruction method can be applied to the data matrix

6 Methods Late-level combination Late-level combination: Supertree Construct separate trees for each gene and combine them to a supertree:

7 Methods Late-level combination Late-level combination: Supertree Construct separate trees for each gene and combine them to a supertree: Supertree methods combine special kinds of information: Split information Matrix Representation: MR with Parsimony (MRP, Baum, 1992; Ragan, 1992) MR with Flipping (MRF, e.g. Chen et al., 2003)

8 Methods Late-level combination Late-level combination: Supertree Construct separate trees for each gene and combine them to a supertree: Supertree methods combine special kinds of information: Triplet information Rooted triplets: MinCut (Semple and Steel, 2000) Modified MinCut (Page, 2002) MaxCut (Snir and Rao, 2006)

9 Methods Medium-level combination Medium-level combination Intermediate data (not final trees) is computed from every source alignment and subsequently combined to a tree. SuperQP: Combination of quartet likelihoods (Schmidt, 2003)

10 Methods Medium-level combination Medium-level combination Intermediate data (not final trees) is computed from every source alignment and subsequently combined to a tree. Average Consensus: Average over distance matrix for each gene (Lapointe and Cucumel, 1997) SDM: Additional weights estimated (Criscuolo et al., 2006)

11 Simulation Simulation setting 1 Estimate an ML tree with branch lengths and model parameters from a data superalignment species tree 2 Generate gene trees 3 Simulate alignments along the gene trees 4 Apply the reconstruction methods to each data set and compare the result with the model tree (true) species tree gene trees (simulated) alignments reconstructed tree

12 Simulation Species tree 10 genes of 25 Crocodylia species (Gatesy et al., 2004) taxa length data sets C_latirostris_5 C_crocodilus_4 M_niger_6 P_palpebrosus_7 P_trigonatus_8 A_mississippiensis_9 A_sinensis_10 O_tetraspis_23 C_cataphractus_22 C_moreletii_14 C_acutus_12 C_intermediu_13 C_rhombifer_11 C_niloticus_21 C_novaeguineae_18 C_mindorensis_17 C_johnstoni_16 C_palustris_20 C_siamensis_15 C_porosus_19 T_schlegelii_24 G_gangeticus_25 Paleognathae_1 Neognathae_2 Testudines_3.10

13 Results Complete and missing data Complete and missing data Step 2: Gene trees are the complete model tree (complete data) or the pruned model tree (missing data) Step 3: Simulation with the parameters estimated with the superalignment 1. Parameters 2. 3.

14 Results Complete and missing data Robinson Foulds distance Complete data Gene Trees MRP MRF ModMinCut MaxCut SuperQP SDM SA

15 Results Complete and missing data Robinson Foulds distance Complete data Gene Trees MRP MRF ModMinCut MaxCut SuperQP SDM SA Robinson Foulds distance Missing data MRP MRF ModMinCut MaxCut SuperQP SDM SA

16 Results Incomplete lineage sorting Incomplete lineage sorting Step 2: For every simulation, a gene tree is generated from the species tree with a coalescent process (θ = 0.005) Step 3: Simulation with the parameters estimated with the superalignment 1. Parameters

17 Results Incomplete lineage sorting Robinson Foulds distance Complete data Gene Trees MRP MRF ModMinCut MaxCut SuperQP SDM SA

18 Results Incomplete lineage sorting Robinson Foulds distance Complete data Gene Trees MRP MRF ModMinCut MaxCut SuperQP SDM SA Robinson Foulds distance Missing data MRP MRF ModMinCut MaxCut SuperQP SDM SA

19 Summary Summary Simulation of sequence-based phylogenetic analysis for multiple data sets With the assumption of tree-like evolution for most genes, superalignment yields the highest accuracy In case of high incongruency among gene trees other methods may outperform superalignment Matrix Representation methods are the best choice for supertree reconstruction

20 Summary Summary Simulation of sequence-based phylogenetic analysis for multiple data sets With the assumption of tree-like evolution for most genes, superalignment yields the highest accuracy In case of high incongruency among gene trees other methods may outperform superalignment Matrix Representation methods are the best choice for supertree reconstruction Acknowledgements: Gregory Ewing (CIBIV) WWTF for funding