Complexity of Splits Reconstruction for Low-Degree Trees

TL;DR

This paper investigates the computational complexity of reconstructing trees from split data, proving NP-completeness in various cases, and offers polynomial algorithms for specific constrained scenarios, with applications in chemical compound design.

Contribution

It establishes NP-completeness for split reconstruction in low-degree trees and provides polynomial algorithms for special cases, advancing understanding of the problem's complexity.

Findings

Reconstruction from splits is NP-complete for paths and certain low-degree trees.

Polynomial algorithms exist when the number of vertex weights or leaves is constant.

The problem relates to chemical compound library design and drug discovery.

Abstract

Given a vertex-weighted tree T, the split of an edge xy in T is min{s_x(xy), s_y(xy)} where s_u(uv) is the sum of all weights of vertices that are closer to u than to v in T. Given a set of weighted vertices V and a multiset of splits S, we consider the problem of constructing a tree on V whose splits correspond to S. The problem is known to be NP-complete, even when all vertices have unit weight and the maximum vertex degree of T is required to be no more than 4. We show that the problem is strongly NP-complete when T is required to be a path, the problem is NP-complete when all vertices have unit weight and the maximum degree of T is required to be no more than 3, and it remains NP-complete when all vertices have unit weight and T is required to be a caterpillar with unbounded hair length and maximum degree at most 3. We also design polynomial time algorithms for the variant where T is required to be a path and the number of distinct vertex weights is constant, and the variant where all vertices have unit weight and T has a constant number of leaves. The latter algorithm is not only polynomial when the number of leaves, k, is a constant, but also fixed-parameter tractable when parameterized by k. Finally, we shortly discuss the problem when the vertex weights are not given but can be freely chosen by an algorithm. The considered problem is related to building libraries of chemical compounds used for drug design and discovery. In these inverse problems, the goal is to generate chemical compounds having desired structural properties, as there is a strong correlation between structural properties, such as the Wiener index, which is closely connected to the considered problem, and biological activity.