Bacterial DNA strand compositional asymmetry: response.

T he main point of this response is to state the primary aim of our paper 1 : our purpose was not to develop yet another method of locating the oriC site based on the GC bias (there are already many excellent methods for this) but to study the effect of this bias on genes. The comparison of our approach with the GC-skew plot, or related techniques, is therefore inappropriate – how can a skew plot predict gene orientation from the protein sequence? It is not a question of us being more (or less) 'objective' as Sam Karlin states; he is referring to a different aim. We would like to address some of his comments specifically. Firstly, he states that our method 'does not use information on variation in intergenic sequences'. Of course it does not, because we were interested in the genes. Nevertheless, it is simple to modify our approach to enable us to work with fragments of chromosomes instead of genes, although an 'in-frame' description would then be impossible. We have performed such experiments to compare the amplitude of the bias in transcribed and non-transcribed regions (E.P.C. Rocha, A. Danchin and A. Viari, unpublished). Interestingly , these experiments have indicated that transcription-coupled repair is not the cause of such bias. Secondly, he states our method 'does not take gene density into ac-count'. Again, this is not an error but a requirement. In GC-skew analysis, the high density of genes in the replicative strand means that the skew is dependent on the global composition of genes, but we wanted to remove this effect to study only the contrast in gene composition (see Ref. 2 for a discussion of this topic). Thirdly, his argument about the covariance matrix is actually related to the question of the optimality of the discrimination (when the condition is not met, one cannot ensure that the discrimination is optimal). However, with a 97%– 99% test-set accuracy (using Bor-relia burgdorferi), one can hardly believe that the discrimination is far from optimal. Of course, this objection could be relevant to other species and we do believe the analysis could be improved by using different discrimination techniques and criteria 3. Lastly, the term 'universal' in the title did not refer to whether or not the bias exists in all bacterial species. It referred to the fact that, when it exists, it always has the same form in terms …