Predicting genetic biodiversity in salamanders using geographic, climatic, and life history traits

or endangered (P. stormi). More recently, Parra Olea et al. [11] discovered five cryptic lineages in Chiropterotriton from Mexico, several of which are threatened due to their restricted ranges [11]. Species with small ranges and/or limited dispersal capabilities can be harder to protect because their distributions often do not fall within protected areas [12] and small ranges are often used as a factor in assigning conservation priorities [13]. Therefore, it is important to identify these hidden lineages, as they could easily go unnoticed and unprotected. Many other species of salamander that would have otherwise gone unnoticed and have been recognized using molecular data have small ranges and likely need protection [14–18]. The presence of cryptic diversity has been recently highlighted as a key component of undescribed biodiversity that requires greater attention [19,20]. Efforts to conserve undescribed genetic diversity can be facilitated using computational methods that identify genetic lineages representing potentially hidden diversity in need of fur- ther investigation. The use of data science techniques has allowed biodiversity studies to expand their geographic and taxonomic focus to explore broader patterns of evolution, which can be difficult to assess using traditional meta-analysis methods [21]. Macrogenetics, a rela- tively new field that merges biodiversity data with genetic data [22,23], has been used to explore how human impacts influence levels of intraspecific genetic diversity [24,25], to study

past and future climate refugia [26,27], and to quantify latitudinal biodiversity gradients [28– 31]. Macrogenetic methods, particularly in combination with predictive modeling, can be used to inform conservation policies by identifying species, taxonomic groups, or geographic areas in need of further investigation [32,33]. Machine learning models excel in processing and analyzing large datasets automatically, making them highly efficient at identifying com- plex, non-linear patterns within extensive data. While these models are powerful tools for making predictions with high accuracy, the process behind many machine learning models can be difficult to interpret [34]. Random forest models, a type of predictive machine learning model, enhance interpretability by providing clear insights into feature importance, thus rep- resenting a powerful tool for identifying factors influencing biodiversity. Recently, such analy- ses have expanded to taxonomic work. Parsons et al. [35] analyzed mitochondrial DNA sequences from over 4000 species of mam- mals, representing roughly 66% of currently described species, and found that mammal diver- sity is largely under-described using molecular species delimitation methods on publicly available barcode data. This is useful for several reasons. A comprehensive list of undescribed genetic lineages that may represent species now exists that can help focus taxonomic efforts. Parsons et al. [35] also found that taxa with small bodies, and large geographic distributions

with variation in precipitation and isothermality, were more likely to contain cryptic diversity. While some of this might seem obvious (morphological differences are harder to observe in small-bodied animals and these animals may be harder to find), it does allow researchers to document characteristics of species, higher taxonomic groups, or even geographic regions that contribute to diversification and therefore biodiversity patterns. When done in disparate taxo- nomic groups (e.g., vertebrates, invertebrates, plants, and fungi) and at different levels (e.g., Class, Order, Family) this furthers our understanding of core evolutionary processes. A similar approach was taken in birds. Using a tree-based molecular species delimitation method, Smith et al. [36] found that latitude explained variation in phylogeographic breaks, while other traits pertaining to habitat and life history explained very little. In this case, phylo- geographic structure was higher in the tropics. Conversely, in other organisms, isolation-by- distance within species is often higher at higher latitudes (multiple taxonomic groups: [32]; amphibians: [37]). Further, genetic variation within amphibians was best explained by range size and elevation, rather than latitude, in the neotropics [37], while latitude was an important predictor of genetic diversity in the nearctic [30]. This suggests that differences exist in how PLOS ONE Predicting hidden lineages in salamanders PLOS ONE | https://doi.org/10.1371/journal.pone.0310932 October 18, 2024 2 / 20 Funding: This work was supported by the

Directorate for Biological Sciences (DEB-1910623 to BCC; DEB-1911293 to TAP). Competing interests: The authors have declared that no competing interests exist.