CREATED: 201004291445 LINK: url:/home/melvin/Modules/Literature/Chen2007.pdf Title: Assessing performane of orthology detection strategies applied to eukaryotic genomes
From the creators of OrthMCL. Methods compared: RIO, Orthostrapper (phylogeny) RSD (distance) BBH, RBH, Inparanoid, OrthoMCL, KOG (BLASTP) SBH, BLASTP, TribeMCL (Homology)
Makes use of latent class analysis (LCA), which is based on the assumption that agreement enhances confidence and disagreement indicates possible errors.
Methods based on phylogeny (RIO, Orthostrapper) or evolutionary distance (RSD) exhibit low FP (1-7%) but high FN (24-64%). Consersely, using homology methods (BLASTP, TribeMCL) results in high FP (50-56%) and low FN (4-5%). Between these two extremes, BLASTP-based orthology prediction methods exhibit a range of FP and FN rates. Only OrthoMCL and Inparanoid have FP and FN < 20%