Coalescent‐based species delimitation meets deep learning: Insights from a highly fragmented cactus system

Abstract

Recognizing species boundaries has long been a major challenge for biologists. The main difficulty is to some degree related to the numerous existing species concepts. The use of specific definitions can lead to alternative strategies for identifying species boundaries in empirical data sets (Carstens et al., 2013;de Queiroz, 2007). However, different species concepts can be considered elements of diverse properties that are associated with the dynamics of the speciation continuum.After the proposal of the unified species concept (Queiroz, 2007), the roles of species concept theory and species delimitation methodologies became apparent. The view of species as independent segments

Evolutionary Genetics of Cacti: Research Biases, Advances and Prospects

“…Consistently, molecular markers have been used to assist species recognition and species circumscription in the cactus family ( Figure 4 ), most of which are biased to PCR-based markers within disparate phylogenetics and population genetics approaches. However, incorporating modern sequencing technologies using coalescent-based methods is of great promise for this purpose [ 107 ].…”

“…The GLCS concept embraces the bridge between microevolutionary processes generating branching patterns, the core of Darwinism theory, and, more specifically, of the phylogeography discipline [ 123 ]. In this sense, the multispecies coalescent model (MSC) is a candidate approach to consistently delimit species according to the GLCS by the statistical modeling of the relationship between the gene trees and the lineage history [ 107 , 124 ]. There are several MSC methods available, including those based on full likelihood (e.g., GMYC [ 125 ]), Bayesian posterior probabilities (BPP) (e.g., [ 126 ]), and supervised [ 127 ] and unsupervised machine learning approaches [ 128 ].…”

“…In our literature survey, we found only one paper using MSC for species delimitation in a cactus taxon. Perez et al [ 107 ] introduced a new framework coupling the MSC with deep learning algorithms (convolutional neural networks) to test alternative species delimitation hypotheses within the P. aurisetus species complex. They were able to select a taxonomic ‘splitter’ hypothesis circumscribing five species instead of the single one currently recognized.…”

Deep Learning in Population Genetics

“…Additionally, we focus our attention on deep neural networks, in all their supervised forms, rather than including other commonly used algorithms such as support vector machine ( Pavlidis et al 2010 ), random forests ( Schrider and Kern 2016 ; Vizzari et al 2020 ), gradient forests ( Laruson et al 2022 ), and hierarchical boosting ( Pybus et al 2015 ). Finally, we restrict our review on applications in population genomics while acknowledging that similar algorithms herein described are used in other related disciplines like genomics ( Yue and Wang 2018 ), phylogenetics ( Suvorov et al 2020 ; Azouri et al 2021 ; Blischak et al 2021 ), phylogeography ( Fonseca et al 2021 ; Perez et al 2022 ), and epidemiology ( Voznica et al 2021 ).…”

dnadna: a deep learning framework for population genetics inference

“…As reviewed by Sanchez et al (2021) , we distinguish two families: those processing many summary statistics, with fully connected or convolutional networks and those based on ’raw’ genetic data leveraging deep architectures to automatically construct informative features (e.g. Adrion et al , 2020b ; Battey et al , 2020 ; 2021 ; Burger et al , 2022 ; Chan et al , 2018 ; Deelder et al , 2021 ; Flagel et al , 2019 ; Fonseca et al , 2021 ; Gower et al , 2021 ; Isildak et al , 2021 ; Meisner and Albrechtsen, 2022 ; Montserrat et al , 2019 ; Perez et al , 2022 ; Qin et al , 2022 ; Sanchez et al , 2020; Torada et al , 2019 ; Wang et al , 2021 ; Yelmen et al , 2021 ). Previous studies have made their implementations available at least for reproducibility and sometimes with a specific effort for re-usability.…”