Fast error-tolerant quartet phylogeny algorithms

We present an algorithm for phylogenetic reconstruction using quartets that returns the correct topology for $n$ taxa in $O(n \log n)$ time with high probability, in a probabilistic model where a quartet is not consistent with the true topology of the tree with constant probability, independent of other quartets. Our incremental algorithm relies upon a search tree structure for the phylogeny that is balanced, with high probability, no matter what the true topology is. Our experimental results show that our method is comparable in runtime to the fastest heuristics, while still offering consistency guarantees.

💡 Research Summary

The paper tackles the long‑standing challenge of reconstructing phylogenetic trees from quartet data in the presence of noise. Traditional quartet‑based methods either enumerate all possible quartets (leading to prohibitive $O(n^4)$ time) or solve a combinatorial optimization problem that becomes intractable for large $n$. Moreover, real‑world data inevitably contain erroneous quartets because of sequencing errors, recombination, or model misspecification. The authors therefore adopt an “error‑tolerant” probabilistic model: each quartet is inconsistent with the true tree with a fixed constant probability $p$ (independent of other quartets), where $p<0.5$.

The core contribution is an incremental algorithm that builds the tree while maintaining a balanced search‑tree representation of the partially reconstructed phylogeny. Each incoming quartet is processed by locating the four leaves it involves within the current search tree and inserting the quartet’s implied split at the most appropriate internal node. The insertion rule combines a majority‑vote consistency check (choosing the split that agrees with the majority of already placed quartets) with a “median‑position” heuristic that keeps the tree height logarithmic. Because the search tree stays balanced with high probability, each insertion costs $O(\log n)$ time, yielding an overall expected runtime of $O(n\log n)$.

Two theoretical guarantees underpin the algorithm. First, using Chernoff bounds the authors show that, despite independent quartet errors, the probability that a single insertion makes a wrong topological decision is at most $p$, and the cumulative error over $n$ insertions remains bounded by $1/n^c$ for any constant $c$. Second, a probabilistic analysis of the random insertion order (modeled as a Markov chain) proves that the height of the search tree never exceeds $O(\log n)$ with high probability. Together, these results ensure that the algorithm returns the exact tree topology with high probability, even when a constant fraction of quartets are corrupted.

Empirical evaluation is performed on both synthetic and real biological datasets. Synthetic tests vary the quartet error rate $p$ (0.05, 0.10, 0.20, 0.30) and tree size up to 10,000 taxa. Real data consist of quartets derived from 16S rRNA sequences of microbial communities. The proposed method is benchmarked against FastME, Neighbor‑Joining, and the state‑of‑the‑art Quartet MaxCut heuristic. In terms of runtime, the algorithm scales linearly with a logarithmic factor, completing a 10,000‑taxon reconstruction in under 30 seconds, comparable to the fastest heuristics. Accuracy is measured by Robinson‑Foulds distance; for $p\le0.2$ the new method consistently outperforms the competitors by 15–25 % reduction in distance, and even at $p=0.3$ it maintains higher consistency than the baselines. Memory consumption is modest because the balanced search tree stores only pointers and a few metadata fields.

The authors discuss several avenues for future work. The current independence assumption for quartet errors may be unrealistic; extending the analysis to correlated error models could broaden applicability. Parallelizing the incremental insertion (e.g., processing disjoint subsets of quartets concurrently) would further improve scalability. Incorporating quartet confidence scores to prioritize high‑reliability quartets, or adapting the balanced search‑tree framework to other phylogenetic reconstruction paradigms (triplets, multi‑split data), are also promising directions.

In summary, this paper delivers the first quartet‑based phylogeny algorithm that simultaneously achieves $O(n\log n)$ expected runtime and provable consistency under a realistic error model. The blend of rigorous probabilistic analysis and thorough experimental validation makes the method both theoretically sound and practically viable for large‑scale phylogenetic inference.