Comparison of complex networks and tree-based methods of phylogenetic analysis and proposal of a bootstrap method

Abstract

Complex networks have been successfully applied to the characterization and modeling of complex systems in several distinct areas of Biological Sciences. Nevertheless, their utilization in phylogenetic analysis still needs to be widely tested, using different molecular data sets and taxonomic groups, and, also, by comparing complex networks approach to current methods in phylogenetic analysis. In this work, we compare all the four main methods of phylogenetic analysis (distance, maximum parsimony, maximum likelihood, and Bayesian) with a complex networks method that has been used to provide a phylogenetic classification based on a large number of protein sequences as those related to the chitin metabolic pathway and ATP-synthase subunits. In order to perform a close comparison to these methods, we selected Basidiomycota fungi as the taxonomic group and used a high-quality, manually curated and characterized database of chitin synthase sequences. This enzymatic protein plays a key role in the synthesis of one of the exclusive features of the fungal cell wall: the presence of chitin. The communities (modules) detected by the complex network method corresponded exactly to the groups retrieved by the phylogenetic inference methods. Additionally, we propose a bootstrap method for the complex network approach. The statistical results we have obtained with this method were also close to those obtained using traditional bootstrap methods.

Determining a cutoff score for the family burden interview schedule using three statistical methods

“…However, cautions also need to be paid during the choice of each method. For tree-based modeling, although it has the advantage of being easy to understand, being useful in data exploration, requiring less data cleaning, with no constraint on data type, and being non-parametric method, it still confronts with the challenge of over fitting, which is one of the most practical difficulties for decision tree models and can only be solved by setting constraints on model parameters and pruning [23–25]. For k-means clustering, its ease of implementation, computational efficiency and low memory consumption has kept it very popular, yet its sensitivity to the initial centroids chosen, the potential bias to create clusters of equal size, and lack of robustness to outliers require further adjustment while using this method [29, 37].…”

“…Having been proposed to be one of the best and mostly used supervised learning methods, tree based methods empower predictive models for both categorical and continuous input and output variables, and map both linear and non-linear relationships quite well [23–25]. Here we use interaction trees to capture treatment-subgroup interactions by recursively splitting the group of patients based on pretreatment characteristics, such that in each split the treatment-split interaction is maximized.…”

Integrating Phylogenetic and Network Approaches to Study Gene Family Evolution: The Case of theAGAMOUSFamily of Floral Genes

“…To summarize the network approach applied here, we describe the main steps performed: The proposed approach used here requires less than 10 seconds to run on an Acer Intel Core i7-6700 CPU @ 3.40GHz for all data sets tested to date (<100 sequences). The scripts used here can be found on GitHub (https:// github.com/deCarvalho90/network_analysis) and the software with a graphical interface is available in the work by Goés-Neto et al 51…”

On the origin of mitochondria: a multilayer network approach

“…Such a multi-dimensional set was employed to simultaneously investigate the structural topology of similarity networks inferred from these proteins, integrating signal from multiple orthologs, all supported by a bootstrap analysis to assess the confidence of the observed clustering patterns. Several contributions by some of us ( Góes-Neto et al, 2010 ; Andrade et al, 2011 ; Carvalho et al, 2015 ; Góes-Neto et al, 2018 , Carvalho, Schnable & Almeida, 2018 ) have shown that the original single-layer network version of this method is reliable and consistent. Indeed, it has been explicitly demonstrated that it produces results comparable to those obtained using distance, maximum parsimony, maximum likelihood, and Bayesian methods ( Góes-Neto et al, 2018 ).…”

“…Nevertheless, broad consensual evidence for its utility in real-world datasets is still lacking ( Spielman & Kosakovsky Pond, 2018 ; Abadi et al, 2019 ; Spielman, 2020 ). As an alternative strategy, several authors (including most of us) have provided evidence that Network Science approaches based on Sequence Similarity Networks (SSNs) ( Bapteste et al, 2013 ) help the discovery of relationships between taxa ( Atkinson et al, 2009 ; Andrade et al, 2011 ; Larremore, Clauset & Buckee, 2013 ; Corel et al, 2016 ; Chowdhary, Löffler & Smith, 2017 ; Solís-Lemus, Bastide & Ané, 2017 ; Góes-Neto et al, 2018 ). Therefore, complex networks have been continuously adopted to explain even complex evolutionary processes, including but not restricted to, horizontal gene transfer, gene domain fusion, and gene or genome introgression ( Corel et al, 2016 ; Pathmanathan et al, 2018 ; Ocana-Pallares et al, 2019 ).…”