MarioniLab/ImmuneAging2017: First release for paper correction

When compiling a vignette describing the statistical analysis underlying Martinez-Jimenez et al., Science, 2017 for inclusion in a Bioconductor workflow associated to the BASiCS software, we noted that the number of spike-in molecules present in the cell lysis volume had been incorrectly calculated, such that all molecule numbers were arbitrarily scaled by the same constant factor. These values are required as an input for the computational analysis implemented in BASiCS. This changes the scale of the arbitrary units in which gene expression is measured, but does not change relative differences between conditions.

Additionally, we noted that using the incorrect number of spike-in molecules led to slightly poorer convergence of the MCMC sampler for lower expressed genes, meaning that our estimates of the mean and dispersion for such genes are less stable. However, in the paper, since we employ a threshold on expression for including genes in downstream analyses, such genes are typically excluded. Moreover, re-running the MCMC sampler with the correct spike-in counts leads to small changes in the set of differentially expressed genes. Importantly, the relative fold changes are highly preserved between the original and the revised analyses.