Mathematical aspects of phylogenetic groves

The inference of new information on the relatedness of species by phylogenetic trees based on DNA data is one of the main challenges of modern biology. But despite all technological advances, DNA sequencing is still a time-consuming and costly process. Therefore, decision criteria would be desirable to decide a priori which data might contribute new information to the supertree which is not explicitly displayed by any input tree. A new concept, so-called groves, to identify taxon sets with the potential to construct such informative supertrees was suggested by An'e et al. in 2009. But the important conjecture that maximal groves can easily be identified in a database remained unproved and was published on the Isaac Newton Institute’s list of open phylogenetic problems. In this paper, we show that the conjecture does not generally hold, but also introduce a new concept, namely 2-overlap groves, which overcomes this problem.

💡 Research Summary

The paper addresses a fundamental problem in phylogenetics: how to decide, before any tree reconstruction, whether a collection of taxon sets can yield new evolutionary information when combined into a supertree. Ané et al. (2009) introduced the notion of a “grove” to capture precisely those collections whose overlap structure guarantees the possibility of resolving at least one cross‑triple—three taxa that are not all contained in any single input set—by some compatible assignment of input trees. A grove is defined so that for every possible partition of its taxon sets either (i) no cross‑triple exists, or (ii) there is an “informative topology assignment” that resolves at least one cross‑triple.

Ané et al. further conjectured that maximal groves in any database enjoy three equivalent properties: (1) the set of maximal groves forms a partition of the whole database, (2) any two intersecting groves have a union that is again a grove, and (3) maximal groves are pairwise disjoint. If true, property (2) would allow a simple algorithmic strategy—identify intersecting groves and merge them—making maximal grove detection computationally tractable.

The present work disproves this conjecture in full generality. After reformulating groves in terms of “split groves” (conditions only on bipartitions) and “tripartition groves” (conditions on three‑way partitions), the authors construct explicit counter‑examples. The most striking example consists of three 2‑element taxon sets S = {{x,y},{y,z},{x,z}}. Every binary split of S lacks a cross‑triple, so S qualifies as a split grove. However, the three‑way partition {{x,y}|{y,z}|{x,z}} creates the cross‑triple {x,y,z}, which cannot be resolved by any topology assignment because each input set contains only two taxa. Consequently, S is not a grove, violating property (2) of the conjecture. Similar counter‑examples persist even when the definition is tightened to require that a grove must contain an informative topology assignment for every partition (the “informative” and “strictly informative” grove variants).

To overcome this limitation, the authors introduce a new class called 2‑overlap groves. A collection of taxon sets is a 2‑overlap grove if any two sets in the collection share at least two taxa. This stronger overlap condition guarantees that any intersecting 2‑overlap groves can be merged without losing the grove property; in other words, property (2) holds for 2‑overlap groves, and therefore the three statements of the original conjecture become true within this restricted framework. The paper proves that under the 2‑overlap condition, every cross‑triple is automatically resolvable by constructing compatible “caterpillar” trees that force a unique resolution of the triple.

Beyond the main result, the paper contributes several auxiliary insights: (i) a clear formalization of cross‑triples, resolved cross‑triples, and informative topology assignments; (ii) a hierarchy of grove concepts (split, tripartition, informative, strictly informative) that clarifies the logical dependencies among the definitions; (iii) lemmas establishing when two taxon sets with at most one shared taxon are always compatible and never produce resolved cross‑triples; and (iv) a constructive method for building informative assignments when the 2‑overlap condition is satisfied.

In summary, the authors demonstrate that the original conjecture about maximal groves is false in the general setting, but they salvage a useful and computationally attractive theory by restricting attention to 2‑overlap groves. This new framework ensures that maximal groves can be identified efficiently and that any union of intersecting groves remains informative, thereby providing a solid mathematical foundation for future large‑scale supertree construction pipelines.