MST-backbone-GM

MST-backbone-GM implements phylogenetic tree inference under the General Markov (GM) model to accommodate non‑stationary evolution such as changes in genomic GC content.

Key Features:

• General Markov (GM) model: Uses the GM model to permit non‑stationary substitution processes and changes in GC content during evolution.
• SEM-GM: Applies a Structural Expectation‑Maximization framework adapted for tree search under the GM model to facilitate efficient tree optimization.
• MST-backbone constraint: Employs a Minimum Spanning Tree framework to constrain the tree search and improve scalability of inference.
• Unrooted inference and rSEM-GM rooting: Infers unrooted trees under the GM model and subsequently roots them using the rSEM-GM procedure.
• Scalability performance: Demonstrates linear scaling with data size for MST-backbone(SEM-GM) (and FastTree) while rSEM-GM, RAxML‑NG, and IQ‑TREE exhibited quadratic scaling in comparative analyses.
• Comparative benchmarking: Evaluated against RAxML‑NG, IQ‑TREE, and FastTree for performance and scaling behavior.
• Empirical validation: Achieved rooting recall rates between 80% and 94% on six empirical datasets under the GM model.
• Limitations for viral data: May fail to realistically root some virus phylogenies, indicating potential overtraining of the GM model on certain empirical datasets.

Scientific Applications:

• Non‑stationary phylogenetic inference: Inferring phylogenies when substitution processes violate stationarity, homogeneity, or time‑reversibility assumptions.
• Modeling GC content evolution: Analyzing datasets where genomic GC content changes over time using a non‑stationary substitution model.
• Rooting phylogenies: Rooting inferred unrooted trees via the rSEM-GM procedure under the GM model.
• Large‑scale dataset analysis: Scalable tree inference for large sequence datasets by constraining search with MST‑backbone.
• Method benchmarking: Comparative evaluation of inference accuracy and scaling behavior against RAxML‑NG, IQ‑TREE, and FastTree.

Methodology:

Implements the General Markov (GM) model, performs tree search with SEM‑GM, constrains search using the MST‑backbone Minimum Spanning Tree framework, and roots trees with rSEM‑GM.