PLoS journal retracts phylogenetics paper

PLoS Computational Biology is retracting a paper published in March that claimed that metrics used to measure the accuracy of phylogenetic trees don't work. Senior author Barry Hall from the Bellingham Research Institute in Bellingham, Washington requested the retraction after a colleague noticed a discrepancy, the fault of a software bug that upended the paper's conclusion. "We are retracting the paper because the conclusion that we came to was completely wrong," Hall told The Scientist. "We found no correlation between clade confidence and phylogenetic tree accuracy, but in fact there is a correlation," Hall said.Once Hall spotted the error, he sent an email to an evolutionary biology mailing list, alerting the community to the error, and saving them from trying to build on his faulty conclusion. "This is the way science should always work," wrote Mike Dunford in a blog entry posted on The Questionable Authority.Currently, scientists looking to test the accuracy of reconstructed phylogenetic trees commonly rely on indirect methods, such as Bayesian posterior probabilities and parametric bootstrap percentages. To study how well these indirect measurements assess phylogenetic tree accuracy, the authors obtained true tree topologies through data simulations and then checked how well phylogenetic reconstructions matched these topologies. "Our conclusions were very striking, and paradigm shifting," coauthor Stephen Salipante, at the University of Washington, told The Scientist. None of the values obtained with the indirect methods showed any correlation with accuracy. After passing peer review, the manuscript was accepted and published. The mistake in the paper was a result of a bug in the software. "Somewhere in the later stages of writing the software, I introduced an error," which "led us to the wrong conclusion," Hall explained. Sarah P. Otto of the University of British Columbia in Vancouver, who was planning to use the paper for a journal club, noticed a discrepancy between an example tree in one of the figures and the results reported for the nuoK gene in a table. She asked to see the nuoK trees using Bayesian probabilities method, and did a manual comparison of those trees with the true trees, after which she concluded that, for that data set, there was a strong correlation between clade confidence -- the reliability of a reconstructed phylogenetic tree -- and the probability of a clade being true. She contacted Hall, who immediately began digging for bugs in the software, and then acknowledged the error. "Barry acted very honorably throughout our email exchange. Both he and I had the spirit of trying to get to the truth," Otto told The Scientist. After informing the PLoS editors about the error, Hall posted a retraction in "Evolution Directory," an evolutionary biology Web site. "It is as important to PLoS Computational Biology as it is to Dr. Hall to correct the record," Catherine Nancarrow, a managing editor at the journal, told The Scientist. "PLoS Computational Biology initiated the formal retraction process immediately upon Dr. Hall's first contact," she said in an Email."As unfortunate as this retraction is for [its authors], it also has brought out the best in academic research," said evolutionary biologist Mario Pineda-Krch at the University of California, Davis, who was not a co-author, in an Email. "The manner in which the retraction was handled by Dr. Hall is commendable, and the sharp intellect and altruistic contribution by Professor Otto, and Dr. Hall's cooperation with her inquiry is admirable and should be applauded," Pineda-Krch added.To Pineda-Krch, the retraction suggests authors should provide open access to source code developed for research purposes, to ensure future bugs are quickly spotted. "In my research, I rely on computer models similar to the one in [this] paper, and I know that despite best intentions and huge efforts, bugs are notoriously difficult to completely eliminate." Graciela Flores mail@the-scientist.comCorrection (posted June 19): When originally posted, the story left out the hyphen in Mario Pineda-Krch's name. We regret the error, which has been corrected.Links within this articleB.G. Hall, S.J. Salipante, "Measures of clade confidence do not correlate with accuracy of phylogenetic trees," PLoS Computational Biology Vol. 3, No. 3, e51 http://compbiol.plosjournals.org/perlserv/?request=get-document&doi=10.1371%2Fjournal.pcbi.0030051Barry G. Hall http://homepage.mac.com/barryghall/BarryHall.html#The Questionable Authority http://scienceblogs.com/authority/2007/06/this_is_the_way_science_should.phpStephen Salipante http://www.gs.washington.edu/academics/gradprogram/salipante.htmSarah P. Otto http://www.zoology.ubc.ca/~otto/Retraction http://evol.mcmaster.ca/~brian/netevoldir/Other/PLoSCompBiol.retractionCatherine Nancarrow http://www.the-scientist.com/article/display/52864A. Gawrylewski, "Your guide to retractions and non-retraction retractions," The Scientist, March 1, 2007. http://www.the-scientist.com/article/display/52864Mario Pineda-Krch http://openwetware.org/wiki/Pineda-KrchA. McCook, "Scientists protest Cell retraction," The Scientist, September 29, 2005. http://www.the-scientist.com/article/display/22782

PLoS journal retracts phylogenetics paper

The Scientist ARCHIVES

Become a Member of

Meet the Author

Graciela Flores

Related Research Resources

What Causes an Earworm?

Research Resources

Podcasts

Webinars

Videos

Infographics

eBooks

Screening 3D Brain Cell Cultures for Drug Discovery

Weaving the Fabric of Cancer Research with Synthetic DNA

Enhancing Elution of Plasmid DNA

Maximizing Lentivirus Recovery

Products

Product News

Sino Biological Sets New Industry Standard with ProPure Endotoxin-Free Proteins made in the USA

Introducing the iQue 5 HTS Platform: Empowering Scientists with Unbeatable Speed and Flexibility for High Throughput Screening by Cytometry

Vanderbilt Selects Parse Biosciences GigaLab to Generate Atlas of Early Neutralizing Antibodies to Measles, Mumps, and Rubella

Shift Bioscience proposes improved ranking system for virtual cell models to accelerate gene target discovery