TRE: Quantifying Tree Reconstruction Error

• TRE is a unit interval metric that measures the topological discrepancy between an original tree and its reconstruction, with TRE = 0 indicating perfect fidelity.
• The methodology uses a stochastic generative model with a tunable parameter (γ) to simulate various tree topologies and controlled sampling-order perturbations to mimic realistic noise.
• Empirical Monte Carlo analysis shows that the sampling disorder probability (p) is the key factor impacting reconstruction accuracy, guiding practical applications in phylogenetics and data mining.

Tree Reconstruction Error (TRE) quantifies the topological discrepancy between an original rooted tree and its reconstruction, particularly under perturbations of the node sampling order. It is formally defined in terms of edge-wise agreement between adjacency structures, using a coincidence similarity index that incorporates both the Jaccard and Interiority measures. TRE is a unit interval metric with $\mathrm{TRE} = 0$ signifying perfect reconstruction and $\mathrm{TRE} \to 1$ indicating maximal dissimilarity. The framework enables rigorous quantification of accuracy loss in applications such as phylogenetics, ontology extraction, and hierarchical data mining when information is accessed or discovered in noisy, out-of-order sequences (Benatti et al., 2022).

1. Generative Model for Rooted Trees

TRE analysis employs a stochastic, single-parameter model to generate rooted trees of size $N$ with continuously tunable "branchiness." Each node $i=1,\dots, n$ at construction is characterized by hierarchical level $h_i$ (measured from the root) and current degree $k_i$ (number of attached children). When adding the $(n+1)$th node, its parent $i$ is chosen at random with probability

$$p_i = \frac{(h_i+1)\, (k_i)^\gamma}{\sum_{j=1}^{n}(h_j+1)\, (k_j)^\gamma}$$

where $\gamma \in \mathbb{R}$ tunes the tree’s topology:

• $\mathrm{TRE} \to 1$0 generates chain-like trees (minimal branching).
• $\mathrm{TRE} \to 1$1 generates bushy, highly branched trees.

The process is iterated until $\mathrm{TRE} \to 1$2. Representative morphologies for $\mathrm{TRE} \to 1$3 are provided to illustrate the continuum from linear to highly branched hierarchies (Benatti et al., 2022).

2. Sampling-Order Perturbations

To simulate realistic reconstruction scenarios, the procedure imposes random perturbations on the canonical node sampling order $\mathrm{TRE} \to 1$4. Each element $\mathrm{TRE} \to 1$5 is independently marked for potential displacement with probability $\mathrm{TRE} \to 1$6, and moved randomly within a window of at most $\mathrm{TRE} \to 1$7 positions of its original index. Parameters:

• $\mathrm{TRE} \to 1$8: the fraction of nodes sampled "out of order."
• $\mathrm{TRE} \to 1$9: maximum positional displacement per shuffled node.

This model directly controls the incidence and severity of sampling-order errors, distinguishing between how frequently nodes are misordered ($N$0) and by how much each can be displaced ($N$1). Such errors mimic realistic acquisition noise in empirical data, permitting study of their impact on topological recovery (Benatti et al., 2022).

3. Coincidence Similarity and Mathematical Formulation of TRE

The error metric compares the original tree $N$2 to a reconstruction $N$3 by flattening their adjacency matrices $N$4 to edge incidence vectors $N$5 ($N$6 possible undirected edges). Three quantitative indices are defined:

• Jaccard index

$N$7

• Interiority index

$N$8

• Coincidence similarity

$N$9

The Tree Reconstruction Error is then defined as: $i=1,\dots, n$0 where $i=1,\dots, n$1 indicates vectorization. This metric robustly penalizes both missing and spurious edges, ensuring that $i=1,\dots, n$2 is sensitive to topological inconsistencies.

4. Empirical Analysis: Monte Carlo Evaluation

Closed-form analytical expressions for $i=1,\dots, n$3 and $i=1,\dots, n$4 are not provided. Instead, extensive Monte Carlo studies parameterize $i=1,\dots, n$5, $i=1,\dots, n$6, and $i=1,\dots, n$7 with $i=1,\dots, n$8, averaging over 30 random trees and 4000 sampling orders per tree: $i=1,\dots, n$9 Findings include:

• $h_i$0 increases monotonically with $h_i$1, showing a sigmoidal response, steepest at low $h_i$2.
• $h_i$3 depends only weakly on $h_i$4 (branchiness) and $h_i$5 (disorder extent).
• $h_i$6 grows with $h_i$7 but is minimally affected by $h_i$8 and $h_i$9.
• The coincidence mode evidences stronger dependence on $k_i$0 and $k_i$1 than the mean.

Representative values at $k_i$2 are $k_i$3 and $k_i$4–0.3. Figures 5–7 of the reference provide supporting statistics and sensitivity plots (Benatti et al., 2022).

5. Observed Impacts of Error Parameters

Key empirical results include:

• Moderate increases in $k_i$5 (up to 4) or changes in tree structure parameter $k_i$6 shift $k_i$7 by only a few percent for fixed $k_i$8.
• Increasing the sampling disorder probability $k_i$9 from $(n+1)$0 to $(n+1)$1 raises $(n+1)$2 from near $(n+1)$3 up to approximately $(n+1)$4–$(n+1)$5.
• The sensitivity $(n+1)$6 is maximal at low $(n+1)$7 (typically $(n+1)$8) and decreases as $(n+1)$9 increases.
• Even a $i$0 rate of out-of-order sampling ($i$1) can reduce edge-wise coincidence by $i$2–$i$3, while subsequent increases in $i$4 yield diminishing effects.

These trends indicate that the dominant determinant of reconstruction fidelity is $i$5, the error probability, rather than the specific tree topology or maximum error extent.

6. Practical Applications and Guidance

The TRE framework provides a precise and operational error metric for hierarchical data systems vulnerable to noisy, non-canonical sampling sequences:

• Systems able to control $i$6 below $i$7 typically achieve above $i$8 edge-wise topological correctness.
• Experimental procedures in phylogenetics, ontology discovery, and tree-based incremental data mining benefit from prioritizing sampling-order reliability over concerns about the precise global branching structure.
• The quantitative calibration of mean and variance in TRE, together with the underlying generative and noise model, facilitates predictive assessment of reconstruction fidelity under empirically realistic error regimes.

Summary points:

• TRE is a mathematically grounded, topological error metric derived from coincidence similarity.
• The single-parameter generative tree model efficiently spans the spectrum from chain to bushy topologies.
• Controlled perturbation models ($i$9, $$p_i = \frac{(h_i+1)\, (k_i)^\gamma}{\sum_{j=1}^{n}(h_j+1)\, (k_j)^\gamma}$$0) realistically emulate imperfect sampling.
• Empirical benchmarks furnish practitioners with actionable expectations for accuracy and variability (Benatti et al., 2022).