Estimating Improved Partitioning Schemes for Ultraconserved Elements

Ultraconserved (UCEs) are popular markers for phylogenomic studies. They are relatively simple to collect from distantly-related organisms, and contain sufficient information to infer relationships at almost all taxonomic levels. Most studies of UCEs use partitioning to account for variation in rates and patterns of molecular evolution among sites, for example by estimating an independent model of molecular evolution for each UCE. However, rates and patterns of molecular evolution vary substantially within as well as between UCEs, suggesting that there may be opportunities to improve how UCEs are partitioned for phylogenetic inference. We propose and evaluate new partitioning methods for phylogenomic studies of UCEs: Sliding-Window Site Characteristics (SWSC), and UCE Site Position (UCESP). The first method uses site characteristics such as entropy, multinomial likelihood, and GC content to generate partitions that account for heterogeneity in rates and patterns of molecular evolution within each UCE. The second method groups together nucleotides that are found in similar physical locations within the UCEs. We examined the new methods with seven published data sets from a variety of taxa. We demonstrate the UCESP method generates partitions that are worse than other strategies used to partition UCE data sets (e.g., one partition per UCE). The SWSC method, particularly when based on site entropies, generates partitions that account for within-UCE heterogeneity and leads to large increases in the model fit. All of the methods, code, and data used in this study, are available from https://github.com/Tagliacollo/PartitionUCE. Simplified code for implementing the best method, the SWSC-EN, is available from https://github.com/Tagliacollo/PFinderUCE-SWSC-EN.