Reconstructing ancient genomes from gene counts: A robust likelihood framework with sampling bias correction.

Area of Science:

• Evolutionary Biology
• Computational Biology
• Genomics

Background:

• Reconstructing ancient genomes is crucial for understanding evolution but is challenged by gene-sequence evolution ambiguities at scale.
• Existing methods for ancestral reconstructions struggle with topological uncertainty in gene trees, limiting their accuracy and scalability.

Purpose of the Study:

• To develop a numerically robust computational framework for accurate ancestral genome reconstruction.
• To overcome the topological uncertainty inherent in gene tree analysis for large-scale genomic datasets.

Main Methods:

• Introduced a phylogenetic gain-loss-duplication (GLD) model based on birth-death processes of gene copies along the species tree.
• Developed efficient computation of likelihood and its gradient, incorporating adjustable observation bias for minimum gene family size.
• Enabled unconstrained numerical likelihood maximization and ancestral inference using posterior probabilities.

Main Results:

• The GLD framework accurately reconstructs ancestral states in a kingdom-level analysis of 269 archaeal genomes.
• GLD resolves artifacts in horizontal gene transfer inferences often seen with methods like ALE, particularly in large alignments.
• Revealed a dual evolutionary mechanism in microbial genomes: tension between streamlining and gene influx, balanced by recurrent losses and massive gains.

Conclusions:

• The GLD framework offers a statistically sound approach for inferring gene content evolution across diverse microbial kingdoms.
• Provides a more accurate understanding of genome evolution by distinguishing genuine evolutionary events from computational artifacts.