Maryam Rabiee Siavash Mirarab Introduction to INSTRAL Species Tree Estimation Adding a species to an existing dataset with a species tree Start from scratch Add directly to species tree ID: 1045051
Download Presentation The PPT/PDF document "INSTRAL: Discordance-Aware Phylogenetic ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. INSTRAL: Discordance-Aware Phylogenetic Placement Using Quartet ScoresMaryam Rabiee, Siavash Mirarab
2. Introduction to INSTRAL
3. Species Tree EstimationAdding a species to an existing dataset with a species treeStart from scratch?Add directly to species tree? Phylogenetic placement
4. Recap: ASTRAL Input: set of gene trees with bipartition set XOutput: species tree that maximizes quartet agreement with gene trees, with all bipartitions from XStatistically consistent under MSCASTRAL-II: adds some bipartitions to X when species are missing from some gene treesASTRAL-III: polynomial time, highly scalable
5. Goal: Quartet Placement Problembackbone species tree on n speciesk gene trees on n+m speciesadd m (query) species to tree
6. Goal: Quartet Placement Problemm = 1compute quartet score (QS) between each possible placement of query taxon on quartet treesQS: number of quartets induced by n+1 species that support that placement
7. Goal: Quartet Placement Problemm > 1: add sequences one after the other
8. Goal: Quartet Placement ProblemORm > 1: add each sequence to backbone and merge
9. INSTRAL Input Takes gene trees on n+m species (n backbone, m query), a species tree on n species, and the labels of the m query speciesTherefore: must have already added the query sequences to the gene treesde novogene tree phylogenetic placement method (e.g., pplacer, APPLES, or EPA-ng)
10. INSTRAL (1 query)Modify ASTRAL algorithmSet bipartition set X to all bipartitions with & without new speciesq: new species; T: backbone tree on leafset L; B(T): all bipartitions of TX includes every possible placement of queryASTRAL guaranteed to find solution to quartet placement problem exactly with new search space
11. INSTRAL (Multiple queries)Independent placementUsed when relationship between queries isn’t important or they are expected to not be closely relatedRuntime increases linearly with number of queriesSpecies tree not statistically consistent (polytomies) Ordered placementRuntime increases as polynomial in number of queriesNOT guaranteed to find optimal quartet placement solutionSpecies tree IS statistically consistent
12. Results
13. DatasetsSimulated Datasets:200 ingroup species50, 200, or 1000 (out of 1000) genes per replicateVarying levels of ILSFastTree-II used to infer gene trees from sequencesTwo inferred species trees (ASTRAL and FastTree-II)Biological Datasets:Insects: 144 species, 1478 genes, ASTRAL species treePlants: 103 species, 424 genes, ASTRAL species treeBirds: 48 orders, full genomes, ASTRAL tree, high gene tree discordance
14. Experiments Leave-one-out: prune one taxon from true/estimated species treeComparison to ASTRAL: Input gene trees are those inferred by FastTree (i.e., de novo)Backbone is pruned estimated species treeReplace taxon with INSTRALCompare the new species tree with the original one inferred by ASTRALSince INSTRAL finds optimal placement, its QS is at least as high as ASTRAL’sComparison to CA-ML (RAxML + EPA-ng): Input gene trees to INSTRAL are either de novo or estimated with EPA-ng to place query sequenceInput to CA-ML is concatenated gene sequencesBackbone is pruned true species treeCompare new species tree with one estimated by EPA-ng
15. Experiments Multiple (ordered) placement: prune a fraction of taxa from estimated species tree, order arbitrarily, and re-insert, updating backbone each timeComparison to ASTRALCompare INSTRAL output species tree with estimated ASTRAL species treeMeasure INSTRAL RF distance to true species tree vs. ASTRAL RF distance to true species treeINSTRAL QS does not have to be at least ASTRAL QS (not guaranteed optimal)
16. Comparison to ASTRALLeft: # cases where INSTRAL QS > ASTRAL QS (ASTRAL failed to find optimal placement) Middle: # cases where INSTRAL RF distance to true tree different from ASTRAL RF distance to true treeRight: # cases where INSTRAL RF distance to true tree < ASTRAL RF distance to true tree50 genes200 genes1000 genesModerate ILS11; 8; 15; 3; 04; 4; 0High ILS41; 31; 1312; 8; 55; 3; 2Very high ILS178; 140; 2641; 33; 719; 12; 6
17. Comparison to CA-ML with EPA-ng EPA-ng finds the maximum likelihood placement of a sequence on a set of candidate branches
18. Multiple (Ordered) PlacementDifference in RF distance between each method (ASTRAL – INSTRAL) and true tree
19. Biological dataSingle placementINSTRAL found the same placement for the query sequence in the species tree as ASTRAL for all casesEPA-ng (on 1KP dataset) only found same placement for 68/103 speciesOrdered placement (remove ½ of species)
20. Phylogenetic Placement on Degraded Multi-Gene Data
21. Phylogenetic Placement on Degraded Multi-Gene DataEvaluating INSTRAL and EPA-ng on genome (multi-gene) data where some of the species have extremely degraded genesMust use EPA-ng to place degraded genes into each gene tree