[Paper Review] Fast phylogeny reconstruction through learning of ancestral sequences

Radu Mihaescu, Cameron Hill+1 more|ArXiv.org|Dec 8, 2008

Algorithms and Data Compression11 references4 citations

TL;DR

This paper presents a fast, O(n³)-time algorithm for phylogeny reconstruction using ancestral sequence learning under the CFN model, achieving reliable sub-forest recovery without prior edge length bounds. It matches the O(log n) sequence length requirement of prior work while significantly improving runtime, enabling deeper tree reconstruction through accurate ancestral inference and minimum spanning tree-based topology inference.

ABSTRACT

Given natural limitations on the length DNA sequences, designing phylogenetic reconstruction methods which are reliable under limited information is a crucial endeavor. There have been two approaches to this problem: reconstructing partial but reliable information about the tree (\cite{Mo07, DMR08,DHJ06,GMS08}), and reaching "deeper" in the tree through reconstruction of ancestral sequences. In the latter category, \cite{DMR06} settled an important conjecture of M.Steel, showing that, under the CFN model of evolution, all trees on $n$ leaves with edge lengths bounded by the Ising model phase transition can be recovered with high probability from genomes of length $O(\log n)$ with a polynomial time algorithm. Their methods had a running time of $O(n^{10})$. Here we enhance our methods from \cite{DHJ06} with the learning of ancestral sequences and provide an algorithm for reconstructing a sub-forest of the tree which is reliable given available data, without requiring a-priori known bounds on the edge lengths of the tree. Our methods are based on an intuitive minimum spanning tree approach and run in $O(n^3)$ time. For the case of full reconstruction of trees with edges under the phase transition, we maintain the same sequence length requirements as \cite{DMR06}, despite the considerably faster running time.

Motivation & Objective

• Address the challenge of reliable phylogeny reconstruction under limited genetic sequence length, especially for deep evolutionary trees.
• Overcome the limitations of existing methods that require prior bounds on edge lengths or fail on deeply divergent topologies.
• Develop a polynomial-time algorithm that reconstructs a reliable sub-forest from short sequences without assuming known edge length constraints.
• Enable deeper tree reconstruction by learning ancestral sequences, improving information flow beyond leaf-level distances.
• Achieve runtime efficiency comparable to standard methods while maintaining the same sequence length requirements as state-of-the-art approaches.

Proposed method

• Use a minimum spanning tree (MST) approach on estimated pairwise distances to infer a sub-forest of the true phylogeny.
• Apply ancestral sequence reconstruction via recursive majority functions to estimate internal node states, enhancing information flow.
• Model the CFN evolutionary process using a percolation framework where edges survive with probability 1−2p(e), and connected components inherit random character values.
• Leverage the Ising model phase transition threshold (λ₀ = log(2)/4) to ensure ancestral sequences can be reconstructed with bounded error.
• Use coupling arguments and symmetry to prove that ancestral state estimates are stochastically more accurate when leaf distances are short.
• Combine ancestral inference with distance estimation to allow reliable topological reconstruction even for deep nodes.

Experimental results

Research questions

• RQ1 Can ancestral sequence learning improve phylogeny reconstruction accuracy under limited sequence length?
• RQ2 Can a polynomial-time algorithm reconstruct a reliable sub-forest without requiring prior bounds on edge lengths?
• RQ3 Does ancestral sequence reconstruction enable deeper tree inference with shorter sequences compared to leaf-only methods?
• RQ4 Can the O(log n) sequence length requirement be maintained while reducing runtime from O(n¹⁰) to O(n³)?
• RQ5 Is the minimum spanning tree approach robust to noise and effective in recovering topological structure from ancestral and leaf sequence data?

Key findings

• The proposed algorithm runs in O(n³) time, a significant improvement over the O(n¹⁰) runtime of prior work under the same model.
• The method achieves full tree reconstruction with high probability when edge lengths are below the Ising model phase transition threshold, matching the O(log n) sequence length requirement of [6].
• Ancestral sequence learning enables reliable reconstruction of deeper nodes by improving information flow beyond leaf-level distances.
• The algorithm constructs a reliable sub-forest even when full tree recovery is impossible, making it robust to data limitations.
• Theoretical analysis confirms that ancestral state estimates are stochastically more accurate when leaf distances are short, supporting the method’s robustness.
• The approach generalizes to group-based models like Kimura 3ST and Jukes-Cantor by reducing them to the binary CFN model via group morphisms to Z₂.