A {Benchmark} of {Parametric} {Methods} for {Horizontal} {Transfers} Detection

Abstract

Horizontal gene transfer (HGT) has appeared to be ofimportance for prokaryotic species evolution. As a consequencenumerous parametric methods, using only the information embeddedin the genomes, have been designed to detect HGTs. Numerousreports of incongruencies in results of the different methodsapplied to the same genomes were published. The use of artificialgenomes in which all HGT parameters are controlled allows testingdifferent methods in the same conditions. The results of thisbenchmark concerning 16 representative parametric methods showed agreat variety of efficiencies. Some methods work very poorlywhatever the type of HGTs and some depend on the conditions or onthe metrics used. The best methods in terms of total errors werethose using tetranucleotides as criterion for the window methodsor those using codon usage for gene based methods and theKullback-Leibler divergence metric. Window methods are verysensitive but less specific and detect badly lone isolated gene.On the other hand gene based methods are often very specific butlack of sensitivity. We propose using two methods in combinationto get the best of each category, a gene based one for specificityand a window based one for sensitivity.