Identifying the footprints of selection in coding sequences can inform about the importance and function of individual sites. Analyses of the ratio of non-synonymous to synonymous substitutions (dN/dS) have been widely used to pinpoint changes in the intensity of selection, but cannot distinguish them from changes in the direction of selection, i.e., changes in the fitness of specific amino acids at a given position. A few methods that rely on amino acid profiles to detect changes in directional selection have been designed, but their performance have not been well characterized. In this paper, we investigate the performance of 6 of these methods.We evaluate them on simulations along empirical phylogenies in which transition events have been annotated, and compare their ability to detect sites that have undergone changes in the direction or intensity of selection to that of a widely used dN/dS approach, codeml’s branch-site model A. We show that all methods have reduced performance in the presence of biased gene conversion but not CpG hypermutability. The best profile method, Pelican, a new implementation of (Tamuri et al., 2009), performs as well as codeml in a range of conditions except for detecting relaxations of selection, and performs better when tree length increases, or in the presence of persistent positive selection. It is fast, enabling genome-scale searches for site-wise changes in the direction of selection associated with phenotypic changes.

The SARS-CoV-2 epidemic in France has focused a lot of attention as it has had one of the largest death tolls in Europe. It provides an opportunity to examine the effect of the lockdown and of other events on the dynamics of the epidemic. In particular, it has been suggested that municipal elections held just before lockdown was ordered may have helped spread the virus. In this manuscript we use Bayesian models of the number of deaths through time to study the epidemic in 13 regions of France. We found that the models accurately predict the number of deaths 2 to 3 weeks in advance, and recover estimates that are in agreement with recent models that rely on a different structure and different input data. In particular, the lockdown reduced the viral reproduction number by ≈ 80%. However, using a mixture model, we found that the lockdown had had different effectiveness depending on the region, and that it had been slightly more effective in decreasing the reproduction number in denser regions. The mixture model predicts that 2.08 (95% CI : 1.85-2.47) million people had been infected by May 11, and that there were 2567 (95% CI : 1781-5182) new infections on May 10. We found no evidence that the reproduction numbers differ between week-ends and week days, and no evidence that the reproduction numbers increased on the election day. Finally, we evaluated counterfactual scenarios showing that ordering the lockdown 1 to 7 days sooner would have resulted in 19% to 76% fewer deaths, but that ordering it 1 to 7 days later would have resulted in 21% to 266% more deaths. Overall, the predictions of the model indicate that holding the elections on March 15 did not have a detectable impact on the total number of deaths, unless it motivated a delay in imposing the lockdown.