Evolving view on phylogenetic networks

summary

The gist

Reticulate processes such as hybridization, introgression, and horizontal gene transfer cannot be fully represented by a bifurcating tree; enter phylogenetic networks…

In short

Phylogenetic networks are needed to model gene flow like hybridization and introgression, which traditional tree models cannot capture. The paper surveys network representations, from split graphs to probabilistic models like the MSNC. It discusses how to infer these complex structures from data while addressing challenges in computational cost and statistical identifiability.

Key concepts

Phylogenetic Networks
These are mathematical structures that go beyond simple branching trees. They include 'hybrid nodes' that represent gene flow events, allowing researchers to visualize reticulate evolutionary histories where species interact through processes like hybridization or horizontal gene transfer.
Level-k Networks
This is a classification system for networks based on their structure. Level-1 networks are the most common and well-studied, while level-k networks allow for more complex reticulation. The paper notes that most current knowledge about inferring these structures comes from the level-1 class.
Multispecies Network Coalescent (MSNC)
This is a probabilistic model that extends the standard species tree model to include gene flow. It allows researchers to define a likelihood function to find the best network structure that fits observed genetic data, incorporating both tree evolution and hybridization events.
Statistical Identifiability
This refers to whether two different networks can produce the exact same pattern of observed data. The paper states that Level-1 semi-directed networks are generally identifiable under the MSNC from gene quartet distributions, meaning they are distinguishable given enough information.

Terminology used across episodes

This episode discusses

The paper

Evolving view on phylogenetic networks · Read on arXiv

Claudia Solís-Lemus

Wisconsin Institute for Discovery, University of Wisconsin-Madison

Reticulate processes such as hybridization, introgression, and horizontal gene transfer cannot be fully represented by a bifurcating tree. Enter phylogenetic networks: first as split graphs to visualize tree discordance, then as explicit probabilistic models that capture biological phenomena. Here, we describe the broad taxonomy of network representations, distinguishing the principal classes of explicit networks, their biological interpretability and our ability to accurately estimate them from empirical data. We also trace the evolution of the main network inferential methods from hybrid detection tests, distance- and subgraph-based amalgamation methods, probabilistic approaches under the multispecies network coalescent, composite-likelihood and divide-and-conquer frameworks, while highlighting the selective pressures of statistical identifiability and computational scalability that have shaped this evolution. As we move towards a network thinking paradigm, previously isolated methodological lineages from population genetics, phylogenomics, and mathematical network theory are now introgressing, uniting diverse network models into a shared framework that can integrate sequence- and species-level reticulate processes, increase robustness to systematic errors, and refine algorithms for genome-scale data, expanding the tree of life into a richer, more entangled yet clearer picture of evolution.

DOI: 10.1093/sysbio/syag045

Transcript

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "Evolving view on phylogenetic networks".

Marcus: Here's the gist of the paper. Reticulate processes such as hybridization, introgression, and horizontal gene transfer cannot be fully represented by a bifurcating tree; enter phylogenetic networks: first as split graphs to visualize tree discordance, then as explicit probabilistic models that capture biological phenomena The paper describes the broad taxonomy of network representations, distinguishing the principal classes of explicit networks, their biological interpretability and our ability to accurately estimate them from empirical data Shifting into the era of network thinking The tree paradigm fails to explain certain patterns (anomalies under the current paradigm) just like any scientific paradigm Gene flow is such an anomaly Historically, gene flow in the form of horizontal gene transfer or hybridization has often been set aside to simplify the reconstruction of the evolutionary histories of species One could argue it is even the norm in some taxonomic groups Gene flow appears at all levels in the tree of life as shown by a large body of evidence

2–11, e.g. [Ines: in recent years] A web of definitions for networks Mathematically, trees are structurally simple At first glance, networks appear to be a straightforward extension of trees by the simple inclusion of a third type of node: hybrid nodes with two parental incoming edges and one outgoing child edge (Figure 1 right) However, one immediate reality complicates our understanding of networks: there is in fact no universal definition of what internal nodes (or edges) represent in what is denoted (quite loosely) a phylogenetic network One such example is haplotype networks

15–17: and their median

18, 19: or median-joining

20: variants which represent the genealogical relationships among distinct haplotypes (unique combinations of linked alleles) observed within a population or species, typically inferred from nonrecombining genetic loci Another such example is implicit or split networks

28–30: , also called abstract or data-display networks Exploring the untamed space of phylogenetic networks Early attempts to make sense of this landscape took an almost taxonomic approach, defining rigorous classes of phylogenetic networks Most methods for inferring networks have specialized in level-1 networks Level-k networks are not necessarily tree child and galled; galled networks are reticulation visible, and all valid networks are stack free Most of what we know about phylogenetic network inference is from the class of level-1 networks Inferring the Network of Life The evolution of network methods has followed an intuitive path Early methods to reconstruct phylogenetic networks started as heuristic or distance-based with limited incorporation of explicit evolutionary models

72: Developments on the mathematical properties of splits, clusters and subgraph patterns within explicit networks gave rise to the formalization of theory and algorithms to build explicit networks from integral subparts Substantial theoretical literature has characterized level-k networks showing, for example, their uniqueness from triplets despite NP-hard reconstruction

99: Networks as parameters in probabilistic models The multispecies coalescent (MSC) model on trees

43, 110, 111: tracks the genealogical history of gene sequences going backwards in time and allows the estimation of species tree topology, divergence times, and population sizes This extended MSC model had one limitation: if there was more than one hybrid species (or multiple individuals sampled from the one hybrid species), all lineages below the hybrid node were forced to follow the same species tree This new probabilistic model that combined ILS and hybridization was denoted indistinguishably the multispecies network coalescent (MSNC) or the network multispecies coalescent (NMSC) and allows us to define a likelihood function to implement a maximum likelihood procedure that will traverse network space in search of the optimum network to fit the data The challenges of network inference Built upon biologically-informed probability models, likelihood-based network inference methods have established as accurate methodologies to infer reticulate evolutionary histories First, the prohibitive computational cost of full likelihood evaluation on phylogenetic networks rendered these methods impractical for large empirical datasets (lack of scalability) Second, as the complexity of the probability models increases, it becomes more likely that different parameter settings produce similar patterns in the data such as gene trees or site frequencies (lack of identifiability) Computational scalability Dividing and conquering FastNet

polynomial complexity; 141: and later InPhyNet

linear complexity; 142: provide a decomposition strategy to split the full problem (network inference on the whole set of taxa) into smaller subproblems (network inferences on a subset of taxa), and then merge solutions of the subproblems into a solution for the full problem Algebraic geometry provides a theoretical framework that enables faster and more efficient phylogenetic network inference through the use of algebraic invariants Statistical identifiability Intuitively, two networks that generate the same data patterns would not be distinguishable from that data alone Level-1 semi-directed networks are generically identifiable under the MSNC from gene quartet distributions except in the presence of 2-cycles and, in certain cases, hybrid nodes and adjacent edges within 3-cycles Generic identifiability means for all values of numerical parameters except some in a set of measure zero (almost all values of numerical parameters) Level-2 networks are not identifiable from average distances, not even generically Interpretations For networks, branch lengths and inheritance probabilities are necessary to provide biological meaning Branch lengths in calendar time could provide information about the time points of the origin and the target of gene flow, and whether these two time points coincide The inheritance probability can also provide information on whether the gene flow event is introgression (small γ) or hybrid speciation (γ close to 0.

Paper summary: Ines: So this paper "Evolving view on phylogenetic networks" looks at how we should be thinking about evolutionary history when gene flow is involved. It claims that just using a simple tree structure isn't enough to capture processes like hybridization or introgression.

Marcus: Exactly. The main point here is moving from tree thinking to network thinking because gene flow, or reticulate evolution, isn't an anomaly anymore; it's actually happening all over the phylogenetic tree of life

two–eleven e.g., in recent years: . This paper sets up the idea that we need these networks to represent those processes better.

Yuki: I wonder what the authors mean by this shift to network thinking? From a population genetic perspective, does this mean we have a better way to trace how gene flow shapes species history?

Ines: It suggests that instead of just seeing one path for evolution, we need something that can show multiple paths and reticulation events. The paper describes the broad taxonomy of these network representations, distinguishing between different classes based on what they are biologically interpretable as and how well we can estimate them from real data <ref:2611.04149#pg1>.

Marcus: That’s a big claim because it moves us past just visualizing discordance with split graphs, which are basically just ways to show where the tree splits differently <ref:2611.04149#pg3>. The paper talks about explicit probabilistic models too, which are supposed to capture the actual biological phenomena <ref:2611.04149#pg1>.

Yuki: So they’re proposing these explicit models as the way forward? What kind of things are we actually trying to capture with these different classes of networks they describe?

Ines: They distinguish between different classes, like level-one networks, and mention that most what we know about inferring them comes from level-one networks <ref:2611.04149#pg2>. It seems they’re focusing on those first because they are the most common ones we can actually work with right now.

Marcus: And the paper traces the evolution of how we infer these networks, starting from hybrid detection tests all the way up to things like divide-and-conquer frameworks <ref:2611.04149#pg2>. It’s a long history of trying to build better tools for this kind of data.

Yuki: I’m curious about the challenges they highlight. They mention statistical identifiability and computational scalability as major issues when you try to fit these complex models to actual empirical data <ref:2611.04149#pg2>.

Ines: Right, that sounds like a huge hurdle because if two different network structures can produce the exact same data patterns, we just can't tell which one is the right one from the data alone <ref:2611.04149#pg2>. And fitting them requires a lot of computational power because it has to account for things like incomplete lineage sorting and gene flow at the same time <ref:2611.04149#pg2>.

Marcus: The paper points out that computational scalability is also an issue, and they introduce strategies like dividing and conquering to handle the problem when you have a whole set of taxa <ref:2611.04149#pg3>. It’s about breaking the big inference problem down into smaller pieces.

Paper summary: Yuki: So for someone looking at this from a population genetics viewpoint, what does it mean that we need these more complex models instead of just tree models? Does it help us understand the real history of species diversification?

Ines: It helps us account for the fact that gene flow is a normal part of evolution, not something to ignore or set aside just to simplify things <ref:2611.04149#pg2>. It forces us to acknowledge how hybridization and introgression can actually influence the patterns we see in genetic data.

Marcus: And they do bring in some concrete terms when talking about what makes these networks identifiable, like generic identifiability under the multispecies network coalescent model <ref:2611.04149#pg2>. That’s a specific mathematical condition for when the network structure is uniquely determined by the data <ref:2611.04149#pg2>.

Yuki: I’m still trying to connect this back to the real world data. What kind of biological meaning are we supposed to get from these networks once we successfully infer them? Are branch lengths and inheritance probabilities really that informative?

Ines: They can be very telling. For instance, branch lengths could give us information about the time points when gene flow happened and whether those events coincided <ref:2611.04149#pg3>. Inheritance probabilities can tell us if the gene flow was more like introgression, which is a smaller amount of gene transfer, or hybrid speciation, which involves a different kind of event <ref:2611.04149#pg3>.

Marcus: That’s where the biological interpretability comes in. The paper stresses that for networks, you need branch lengths and inheritance probabilities to give them meaning beyond just being a visual structure <ref:2611.04149#pg3>. It’s about adding context to the shape of the network.

Yuki: It sounds like this research is pushing us toward a more nuanced picture of evolutionary history, where we can actually incorporate these reticulate events into our models, rather than just trying to fit them onto a simpler tree structure <ref:2611.04149#pg2>.

Ines: Right. The paper's title "Evolving view on phylogenetic networks" really captures that idea—that our understanding needs to evolve beyond the basic bifurcating tree model to truly represent biological reality <ref:2611.04149#pg1>.

Marcus: And the authors are doing a lot of groundwork by tracing this whole methodological path, from early hybrid detection tests to these more advanced probabilistic models under the multispecies network coalescent <ref:2611.04149#pg2>. It’s a comprehensive look at where we came from in this area.

Yuki: It seems like the paper isn't just presenting one new technique, but rather mapping out the entire landscape of how researchers approach inferring these complex evolutionary histories <ref:2611.04149#pg2>.

Ines: And it clearly lays out what the big challenges are—identifiability and computation—so we know where the current limitations of this field lie <ref:2611.04149#pg2>.

Marcus: So, ultimately, this paper is about giving us a clearer taxonomy of networks so we can better decide which mathematical framework to use for our specific biological questions <ref:2611.04149#pg3>.

Yuki: It’s about recognizing that the history of life isn't always a simple branching tree, and acknowledging the complexity of gene flow is necessary for an accurate picture <ref:2611.04149#pg2>.

Conclusion: Ines: So we’ve been looking at how we move from simple tree shapes to these more complex network structures, and now we're getting to the conclusion of this paper, "Evolving view on phylogenetic networks."

Marcus: Yeah, so what’s the big picture here? What are the main points about this work by

insert authors if known or just say 'the authors': ?

Ines: Basically, they’re saying we have to stop thinking only in terms of simple branching trees when we look at evolution. It’s about accepting that gene flow is a normal part of how life happens, and those processes create these reticulations that a tree just can't capture.

Marcus: And what does that mean for the actual data? Are they saying this new way of thinking lets us actually get better at inferring those evolutionary histories than before?

Ines: The paper argues that using explicit probabilistic models, like the network multispecies coalescent model, is how we can actually define a likelihood function to find the best network structure that fits what we see in the genetic data.

Marcus: That sounds heavy computationally. What are the practical implications for someone working with huge cohorts or big datasets? Is it just theoretical now?

Ines: It’s not just theory. They show that even with those computational hurdles, methods like dividing and conquering can help make these inferences scalable enough to handle larger sets of taxa than we could before.

Marcus: So the conclusion is that these explicit networks aren't just a fancy way to draw a picture of discordance; they are the actual mathematical tools we need for modeling complex evolution.

Ines: Exactly. It shifts our focus from just visualizing weird splits to actually using models that can explain *why* those reticulations happen in biological terms.

Marcus: It gives us a clearer taxonomy, which means when I look at my data, I have a better framework to decide if I'm looking at an introgression event or something else entirely.

Ines: Right. And that brings us to the next thing—how do we actually use these inferred networks to tell the real story about species diversification?

More episodes

← Home