Abstract
Analyzing past ecosystems can improve our understanding of the mechanisms linking biodiversity with environmental changes. Sedimentary ancient DNA (sedaDNA) opens a window to past biodiversity, beyond the fossil record, that can be used to reconstruct ancient environments and ecosystems functions. To this end, modern biodiversity and environmental conditions are used to calibrate transfer functions, that are then applied to past biodiversity data to reconstruct environmental parameters. Doing this with sedaDNA can be challenging, because ancient DNA is often obtained in limited quantities and fragmented into smaller molecules. This leads to noisy datasets, with a low alpha diversity relative to modern DNA, patchy taxa detection patterns and/or skewed relative abundance profiles. How this affects beta-diversity measures, and the performance of transfer functions remain untested. Here we simulated ancient DNA reads counts matrices from synthetic and empirical datasets, and tested 464 combinations of counts transformations (n = 13), beta-diversity indices (n = 16), and ordinations methods (n = 4), and assessed their performance in (i) separating the ecological signal from the noise introduced by DNA degradation and in (ii) predicting ground-truth environmental conditions. Our results show that commonly used workflows in DNA-based community ecology studies are sensitive to the noise associated to ancient DNA signal. Instead, combinations of methods that include more recent ordination methods proved robust to ancient DNA noise and produced better transfer functions. Our study provides a framework for designing postprocessing workflows that are better suited for sedaDNA studies.