Strand orientation, rate heterogeneity and rearrangements: a check of an alignment-free test for recombination in bacterial genomes

A replication and check of the alignment-free test for recombination of Haubold et al. (2013, Bioinformatics 29:3121–3127) on the 58 Escherichia coli genomes of the original study, with a census of strand orientation in 720 complete bacterial genomes from 80 genera. Main findings. (1) The test’s statistic Dr is far from its assumed standard normal distribution even when its model holds, and a mutation rate that varies along the genome, without recombination, makes 95–100% of simulated pairs significant. (2) Nine of the 58 genomes are deposited as the reverse complement of the others. Because the method compares one strand only, all fifteen pairs with the highest variance ratio are pairs in opposite orientation, and the one pair with a known transfer (KO11 vs K-12 MG1655) moves from rank 481 to rank 2 after orientation. In the census, 24% of genomes are deposited in reverse relative to the reference of their species and 23% match neither orientation as a whole. The published NCBI submission rules set no convention for the strand of a complete chromosome. (3) An order-dependent scan with a block-resampling null, tested as a remedy, is calibrated only when rate variation is shorter than the block; on near-identical genomes it reports rearrangements, such as the inversion between rRNA operons in K-12 W3110. Practical consequence: any comparison of genomes that reads one strand only should orient the genomes first and report the orientation used; genomes that match neither orientation need a structural comparison. Files: the preprint (PDF) and four CSV tables with a README describing every column. All genomes are public NCBI records. The study was carried out with the assistance of an AI system (Claude, by Anthropic); see the preface.