Finiteness results for abelian tree models

Commutative Noetherian rings and modules (13E05) Applications of commutative algebra (e.g., to statistics, control theory, optimization, etc.) (13P25) Determinantal varieties (14M12) Multilinear algebra, tensor calculus (15A69) Applications of statistics to biology and medical sciences; meta analysis (62P10)

Abstract: Equivariant tree models are statistical models used in the reconstruction of phylogenetic trees from genetic data. Here equivariant refers to a symmetry group imposed on the root distribution and on the transition matrices in the model. We prove that if that symmetry group is Abelian, then the Zariski closures of these models are defined by polynomial equations of bounded degree, independent of the tree. Moreover, we show that there exists a polynomial-time membership test for that Zariski closure. This generalises earlier results on tensors of bounded rank, which correspond to the case where the group is trivial, and implies a qualitative variant of a quantitative conjecture by Sturmfels and Sullivant in the case where the group and the alphabet coincide. Our proofs exploit the symmetries of an infinite-dimensional projective limit of Abelian star models.

Review text

This article studies equivariant tree models on evolutionary trees. An equivariant model is specified by a rooted tree \(T\), a finite alphabet \(B\) and a group \(G\) acting on \(B\). There is a random variable \(X_v\) associated to each vertex \(v \in \text{vert}(T)\). These random variables take values in the alphabet \(B\). The model parameters are the distribution \(\pi\) at the root \(r\) and transition matrices \(A_e\) for each edge \(e \in \text{edge}(T)\). The transition matrices satisfy \(A_e(gb,gb') = A_e(b,b')\) for all \(g \in G\) and \(b,b' \in B\). The probability of observing a word \(\mathbf{b}\) at the leaves of \(T\) is \[ P(\mathbf{b}) = \sum_{\mathbf{b}' \in B^{\text{vert(T)}} \text{ extending } \mathbf{b}} \pi(b'_r) \prod_{e=(v,v') \in \text{edge}(T)} A_e(b'_v,b'_{v'}). \] An equivariant model is the set of joint probability distributions in \(\mathbb{R}^{B^{\text{leaf}(T)}}\) for all parameter values for fixed \(T, B\) and \(G\). The first main result of this paper states that if \(G\) is an abelian group, then there exists a bound \(D(B,G)\) such that the Zariski closure of the equivariant model for any \(T\) and given \(B\) and \(G\) is defined by polynomial equations of degree at most \(D(B,G)\). If \(G\) is abelian, \(B=G\) and \(G\) is acting by left multiplication on itself, then the resulting model is a group-based model and this Zariski closure of the model is a toric variety. It was conjectured by \textit{B. Sturmfels} and \textit{S. Sullivant} [J. Comp. Biol. 12, 204--228 (2005; Zbl 1391.13058)] that the toric ideal of the toric variety is generated by polynomials of degree at most \(|G|\). The original conjecture was proven for \(G=\mathbb{Z}_2\) by Sturmfels and Sullivant and for \(G=\mathbb{Z}_2 \times \mathbb{Z}_2\) by \textit{M. Michałek} and \textit{E. Ventura} [``Phylogenetic complexity of the Kimura 3-parameter model, \url{arXiv:1704.02584}]. The result in this paper is the first general result for any abelian group \(G\) towards the conjecture by Sturmfels and Sullivant. The difference from the conjecture is that the result in this paper does not give an explicit bound and it is for the generators of the radical of the ideal (this is a set-theoretic result whereas the original conjecture is ideal-theoretic). The second main result of the paper states that there exists an algorithm that decides in polynomial time whether a joint probability vector lies in the Zariski closure of the equivariant model for \(T,B\) and \(G\), where \(G\) is an abelian group. This proof is nonconstructive and it does not give an explicit algorithm.

Recommendations

Cites work

Cited in

(14)

This page was built for publication: Finiteness results for abelian tree models

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2344494)