Subtree power analysis finds optimal species for comparative genomics

Jon McAuliffe, Michael I. Jordan, Lior PachterarXiv 2004 · DOI · arXiv

Abstract

Sequence comparison across multiple organisms aids in the detection of regions under selection. However, resource limitations require a prioritization of genomes to be sequenced. This prioritization should be grounded in two considerations: the lineal scope encompassing the biological phenomena of interest, and the optimal species within that scope for detecting functional elements. We introduce a statistical framework for optimal species subset selection, based on maximizing power to detect conserved sites. In a study of vertebrate species, we show that the optimal species subset is not in general the most evolutionarily diverged subset. Our results suggest that marsupials are prime sequencing candidates.

Introduction

Comparative genomics is built on the principle that functional genomic regions—such as protein-coding genes or regulatory elements—evolve more slowly than non-functional "junk" DNA due to purifying selection. By aligning sequences across diverse species, researchers can identify these highly conserved regions as candidates for biological importance.

However, sequencing remains a resource-intensive endeavor, necessitating a strategic approach to selecting which genomes to prioritize. Historically, the prevailing heuristic has been to maximize "additive divergence," or selecting species that are as evolutionarily distant as possible. This paper challenges that convention, demonstrating that the most divergent species are often not the most informative. Instead, the authors propose a rigorous statistical framework to identify the "Goldilocks zone" of evolutionary distance.

The Limits of Additive Divergence

The fundamental flaw in the "most divergent" strategy lies in the signal-to-noise ratio. If the selected species are too closely related, there has been insufficient time for neutral mutations to accumulate, making it impossible to distinguish functional conservation from recent common ancestry. Conversely, if the species are too distant, the "signal" of conservation is lost to random noise; multiple mutations at the same site (homoplasy) can make a conserved site look neutral, or vice versa.

To address this, the authors frame species selection as a decision-theoretic problem. The goal is to maximize statistical power—the probability of correctly identifying a conserved site—while maintaining a fixed false discovery rate (typically 5%).

Methodology: A Power Analysis Approach

The researchers utilized Neyman-Pearson hypothesis testing to evaluate different species subsets. They analyzed two primary scenarios:

  1. FOSST (Fully Observed): A theoretical model where the ancestral DNA sequence is known.
  2. HASST (Hidden Ancestor): A realistic model where the ancestral sequence is unknown and must be inferred.

Central to their method is the concept of the $k$-MPSS ($k$-most-powerful Steiner subtree). This represents the specific subset of $k$ species that provides the highest statistical power for a given evolutionary model. The authors employed the Jukes-Cantor substitution model to describe DNA base mutations and validated their framework using a well-characterized genomic region: the Cystic Fibrosis Transmembrane conductance Regulator (CFTR) across 21 representative vertebrates.

Key Findings

1. The Divergence "Sweet Spot"

The study reveals that statistical power does not increase linearly with evolutionary distance. Instead, there is a unique, finite branch length that maximizes power. Beyond this point, the accumulation of random mutations degrades the ability to detect selection. Consequently, the subset of species with the most total evolutionary distance (the $k$-MDSS) frequently differs from the subset that provides the most statistical signal (the $k$-MPSS).

2. The Marsupial Advantage

In the vertebrate study, the framework identified a surprising priority. While the platypus is more evolutionarily diverged from humans than marsupials are, adding a marsupial like the dunnart to the existing pool of sequenced genomes provided a significantly larger power boost. Specifically, the dunnart improved the power to detect conserved elements by a relative 12.5%, outperforming the more distant platypus.

3. Quantifying Power Loss

The risks of relying on the divergence heuristic are substantial. In simulations involving high mutation rates, the "most divergent" four-species subset shared only one species with the "most powerful" subset. Relying on the divergent subset resulted in an 8.5% absolute loss in power to detect conserved sites.

Discussion and Limitations

While this framework provides a more robust foundation for species selection, the authors note several areas for future refinement:

  • Single-Nucleotide Focus: The current model calculates power for individual DNA bases. In practice, researchers often look at multi-site windows, which would likely increase the overall power and potentially shift the optimal distances.
  • Model Sensitivity: The results are dependent on the accuracy of the underlying phylogenetic tree and the Jukes-Cantor mutation model. Significant errors in the estimated tree topology or mutation rates could lead to sub-optimal species choices.
  • Computational Complexity: Identifying the $k$-MPSS is a combinatorial optimization problem. While feasible for dozens of species, it becomes computationally demanding as the pool of candidate species grows into the thousands.

Conclusion

This research demonstrates that more data is not always better data. By shifting the focus from maximizing divergence to maximizing statistical power, researchers can more efficiently allocate sequencing resources. The identification of marsupials as high-priority candidates highlights the practical utility of this framework in guiding the next generation of comparative genomic studies.

Summary updated

Abstract

Sequence comparison across multiple organisms aids in the detection of regions under selection. However, resource limitations require a prioritization of genomes to be sequenced. This prioritization should be grounded in two considerations: the lineal scope encompassing the biological phenomena of interest, and the optimal species within that scope for detecting functional elements. We introduce a statistical framework for optimal species subset selection, based on maximizing power to detect conserved sites. In a study of vertebrate species, we show that the optimal species subset is not in general the most evolutionarily diverged subset. Our results suggest that marsupials are prime sequencing candidates.
BibTeX
@misc{mcauliffe2004subtree,
  author = {Jon McAuliffe and Michael I. Jordan and Lior Pachter},
  title = {{Subtree power analysis finds optimal species for comparative genomics}},
  year = {2004},
  eprint = {q-bio/0412012},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/q-bio/0412012},
}
Figure