Statistical Distributions of Genome Assemblies Reveal Random Effects in Ancient Viral DNA Reconstructions

Ancient human viruses have been detected in ancient DNA (aDNA) samples of both Anatomically Modern Humans and Neanderthals. Reconstructing genomes from aDNA using reference mapping presents numerous problems due to the unique nature of ancient samples, their degraded state, smaller read sizes and the limitations of current methodologies. The spurious alignments of reads to reference sequences (mapping) are a main source of false positives in aDNA assemblies and the assessment of signal-to-noise ratios is essential to differentiate bona fide reconstructions from random, noisy assemblies. Here, we analyzed the statistical distributions of viral genome assemblies, ancient and modern, and their respective random "mock" controls used to evaluate the signal-to-noise ratio. We tested if differences between real and random assemblies could be detected from their statistical distributions. Our analysis shows that the coverage distributions of (1) real viral aDNA assemblies of adenovirus (ADV), herpesvirus (HSV) and papillomavirus (HPV) do not follow power laws nor log-normal laws, (2) (ADV) and control aDNA assemblies are well approximated by log-normal laws, (3) negative control parvovirus B19 (real and random) follow a power law with infinite variance and (4) the mapDamage negative control with non-ancient DNA (modern ADV) and the mapDamage positive control (human mtDNA) are well approximated by the negative binomial distribution, consistent with the Lander-Waterman model. Our results show that the tails of the distributions of aDNA and their controls reveal the weight of random effects and can differentiate spurious assemblies, or false positives, from bona fide assemblies.