Generalized Fitch Graphs II: Sets of Binary Relations that are explained by Edge-labeled Trees

TL;DR

This paper generalizes Fitch graphs to sets of binary relations explained by edge-labeled trees with subsets of colors, providing characterizations, uniqueness results, and polynomial algorithms for recognition and construction.

Contribution

It introduces a generalized framework for Fitch graphs with multi-colored edge labels, characterizes them, and offers efficient algorithms for their recognition and least-resolved tree construction.

Findings

Characterization of Fitch maps via neighborhoods and forbidden submaps

Proof of uniqueness of the least-resolved explaining tree

Polynomial-time algorithm for recognition and construction

Abstract

Fitch graphs $G=(X,E)$ are digraphs that are explained by $\{\emptyset, 1\}$-edge-labeled rooted trees $T$ with leaf set $X$: there is an arc $(x,y) \in E$ if and only if the unique path in $T$ that connects the last common ancestor $\mathrm{lca}(x,y)$ of $x$ and $y$ with $y$ contains at least one edge with label "1". In practice, Fitch graphs represent xenology relations, i.e., pairs of genes $x$ and $y$ for which a horizontal gene transfer happened along the path from $\mathrm{lca}(x,y)$ to $y$. In this contribution, we generalize the concept of Fitch graphs and consider trees $T$ that are equipped with edge-labeling $\lambda: E\to \mathcal{P}(M)$ that assigns to each edge a subset $M'\subseteq M$ of colors. Given such a tree, we can derive a map $\varepsilon_{(T,\lambda)}$ (or equivalently a set of not necessarily disjoint binary relations), such that $i\in \varepsilon_{(T,\lambda)}(x,y)$ (or equivalently $(x,y)\in R_i$) with $x,y\in X$, if and only if there is at least one edge with color $i$ from $\mathrm{lca}(x,y)$ to $y$. The central question considered here: Is a given map $\varepsilon$ a Fitch map, i.e., is there there an edge-labeled tree $(T,\lambda)$ with $\varepsilon_{(T,\lambda)} = \varepsilon$, and thus explains $\varepsilon$? Here, we provide a characterization of Fitch maps in terms of certain neighborhoods and forbidden submaps. Further restrictions of Fitch maps are considered. Moreover, we show that the least-resolved tree explaining a Fitch map is unique (up to isomorphism). In addition, we provide a polynomial-time algorithm to decide whether $\varepsilon$ is a Fitch map and, in the affirmative case, to construct the (up to isomorphism) unique least-resolved tree $(T^*,\lambda^*)$ that explains $\varepsilon$.