Course Hero Logo

Running time we ran ca ml and astral ii on different

Course Hero uses AI to attempt to automatically extract content from documents to surface to you and others so you can study better, e.g., in search results, to enrich docs, and more. This preview shows page 78 - 81 out of 122 pages.

Running time:We ran CA-ML and ASTRAL-II on different platforms, andhence cannot make direct running time comparisons. Nevertheless, we provideour running time numbers to give a general idea. CA-ML using FastTree on200-taxon model conditions with 1000 genes took roughly two hours, whereasASTRAL-II took roughly one hour to estimate the species tree, and estimatinggene trees also took about 1.5 hours. In general, therefore, the running timesof ASTRAL-II and CA-ML are relatively close on this dataset.68
10M2M500K0.00.10.20.30.40.00.10.20.30.41e-061e-07502001000502001000502001000genesSpecies tree topological error (RF)ASTRAL-IIASTRAL-II (true gt)CA-MLFigure 5.16:Comparison of ASTRAL-II run on estimated and truegene trees and CA-ML on Dataset I. The different between ASTRAL-IIwith true gene tree (“true gt”) and ASTRAL-II with estimated gene trees in-dicates the impact of gene tree error. Note that with true gene trees, ASTRALhas excellent accuracy and is always better than CA-ML (using FastTree).5.4.2.4RQ4: Effect of gene tree errorIn RQ3, we observed that under some conditions, CA-ML was moreaccurate than ASTRAL-II, a pattern that we attribute to high levels of genetree error present in our simulations.When true (simulated) gene trees areused instead of the estimated gene trees, the accuracy of ASTRAL-II is out-standing, regardless of the model condition (see Fig. 5.16) and ASTRAL-II isalways more accurate than CA-ML. Thus, the fact that CA-ML is occasion-ally more accurate than ASTRAL-II under lower levels of ILS is related toestimation error in the input provided to ASTRAL-II.69
In our ASTRAL-II and NJst analyses, gene tree error had a positivecorrelation with species tree error (Fig. 5.17), with correlation coefficients thatwere similar for ASTRAL-II and NJst. The error of CA-ML also correlatedwith gene tree error (obviously the relationship is indirect as factors such asshort alignments impact both CA-ML and gene tree error), but the correla-tion was weaker than the correlation observed for coalescent-based methods(Fig 5.18).Interestingly, the correlation between gene tree estimation errorand species tree error was typically higher with fewer genes.To further investigate the impact of the gene tree error, we dividedreplicates of each model condition into three categories:average gene treeestimation error below 0.25 is labelled low, between 0.25 and 0.4 is labelledmedium, and above 0.4 is labelled high.We plotted the species tree errorwithin each of these categories (see Figs. 5.19 and 5.20).The relative per-formance of ASTRAL-II and NJst is typically unchanged across various cat-egories of gene tree error, but increasing gene tree error tends to increases inthe magnitude of the difference between ASTRAL-II and NJst. Furthermore,MP-EST seemed to be more sensitive to gene tree error than either NJst orASTRAL-II (Fig. 5.20).

Upload your study docs or become a

Course Hero member to access this document

Upload your study docs or become a

Course Hero member to access this document

End of preview. Want to read all 122 pages?

Upload your study docs or become a

Course Hero member to access this document

Term
Fall
Professor
NoProfessor
Tags

Newly uploaded documents

Show More

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture