A Simulation Study of the Reliability of Recombination Detection Methods

Abstract

There exist many methods to detect recombination or mosaic structure in a sample of DNA sequences. But how reliable are they? Four methods were investigated with respect to their power to detect recombination in simulated samples with different amounts of recombination and mutation. In addition, we investigated the impact of the shape of the underlying genealogy on their performances. We found that the methods detected far fewer recombinations than were theoretically possible and that methods based on the principle of incompatibility in general had more power than methods that did not make use of this principle explicitly. This seemed, in particular, to be the case for phylogenies generated under population expansion scenarios which result in long branches at the tips and small deep branches. In addition to the results obtained through simulations, a series of new theoretical results on recombination is presented.

A Revised Evolutionary History of the CYP1A Subfamily: Gene Duplication, Gene Conversion, and Positive Selection

“…However, it is widely recognized that recombination detection algorithms tend to underestimate the incidence of gene conversion in real datasets (Posada and Crandall, 2001;Wiuf et al, 2001;Posada, 2002). In the current case, the concordance of SlidingBayes and RDP2 results and the presence of corroborating stretches of paralog-paralog sequence identity provide a high degree of confidence in predicted gene conversion tracts, even those detected by only one of five statistical algorithms.…”

“…Phylogenetic methods are generally considered to be the least sensitive means of recombination detection, but may be most appropriate in cases of frequent gene conversion (Drouin et al, 1999;Posada and Crandall, 2001;Posada, 2002). In contrast, many statistical algorithms lose sensitivity when gene conversion is common and sequence divergence is low (Posada and Crandall, 2001;Wiuf et al, 2001;Posada, 2002). Conversely, those same algorithms may be subject to false positives when presented with highly divergent datasets.…”

Computational Methods for Analysis of Cryptic Recombination in the Performance of Genomic Recombination Detection Software

“…Methods work by identifying discontinuities in sequence similarities or genetic distance (e.g., GENECONV [6,7] and MAXCHI [4,8]) or by phylogenetic methods that, for example, identify incongruous tree topologies (e.g., RECPARS [1]). Many methods have been evaluated [4,5,11] and new ones continue to be developed. In comparisons of recombination analysis software, GENECONV was among the most highly ranked [4] and is included in the RDP2 analysis suite [2].…”

An Exact Nonparametric Method for Inferring Mosaic Structure in Sequence Triplets

“…Detecting recombination usually involves searching groups of sequences for candidate recombinants or recombination signals and testing whether these represent statistically significant departures from expectation under a null hypothesis of no recombination. Dozens of statistical tests have been developed (Stephens 1985;Sawyer 1989;Balding et al 1992;Karlin and Brendel 1992;Maynard Smith 1992;Takahata 1994;Sneath 1995;Goss and Lewontin 1996;Jakobsen and Easteal 1996;Grassly and Holmes 1997;Maynard Smith and Smith 1998;Sneath 1998;Awadalla et al 1999;Crandall and Templeton 1999;Holmes et al 1999;Wall 1999;Gibbs et al 2000;Martin and Rybicki 2000;Worobey 2001;Bruen et al 2006) and evaluated (Wall 2000;Brown et al 2001;Posada and Crandall 2001;Wiuf et al 2001;Posada 2002) in this endeavor, none of which has yet emerged as the single standard test to be used for identifying recombination. In addition to testing for the existence of recombination, certain methods are also able to locate recombination breakpoints and, sometimes, the parent sequences involved in the recombination event, although the latter can be quite difficult.…”