Ancestral Reconstruction - Usage Guide

Overview

Ancestral state reconstruction (ASR) infers states at internal phylogenetic nodes for sequences (codon, protein, nucleotide), discrete traits, continuous traits, and gene content. The choice of framework depends on the data class and inference question. For sequences, PAML codeml RateAncestor=1 (Yang 1995 Genetics 141:1641) and IQ-TREE2 --ancestral provide marginal ML reconstructions; GRASP (Foley 2022) handles indels probabilistically -- critical for protein resurrection. For discrete traits, corHMM (Boyko & Beaulieu 2021) with hidden Markov rate models supersedes simpler Mk approaches.

The most consequential mistake is reconstructing under a site-independent or trait-stationary model when the biology demands a hidden-rate / non-stationary / epistatic model -- the resulting "ancestor" is mathematically optimal under the wrong model and silently wrong.

Prerequisites

# Sequence ASR
conda install -c bioconda paml iqtree

# RevBayes for Bayesian ASR
# https://revbayes.github.io/download

# GRASP for indel-aware protein ASR
# https://github.com/bodenlab/GRASP

# FastML
conda install -c bioconda fastml

# R packages for trait ASR
Rscript -e "install.packages(c('ape', 'phytools', 'geiger', 'corHMM', 'phangorn', 'OUwie', 'bayou', 'RERconverge'))"
Rscript -e "remotes::install_github('thej022214/hisse')"

# Python
pip install biopython ete4

Quick Start

Tell the AI agent what to do: - "Reconstruct the ancestral protein sequence at the mammalian root for resurrection construct design" - "Infer ancestral state of trait X using corHMM with hidden rate categories" - "Perform stochastic character mapping with 1000 simulations to get full posterior over character histories" - "Reconstruct ancestral continuous trait values along the tree under OU model with regime shifts"

Example Prompts

Protein Resurrection

"Reconstruct the ancestral cytochrome c sequence at the eukaryote root using GRASP for indel-aware ASR. Design 8 alternative constructs varying ambiguous sites (P < 0.8). Validate ML construct against extant homologs as functional baseline."

Discrete Trait Evolution

"Reconstruct ancestral states of a binary trait (flight ability) on 50 bird species. Test ER vs SYM vs ARD rate models via AIC. If rates are asymmetric, use ARD. Cross-validate against stochastic mapping with phytools::make.simmap(nsim=5000, pi='fitzjohn')."

State-Dependent Diversification

"Test whether a discrete trait (sociality) influences speciation/extinction across 200 species using HiSSE as the required null. BiSSE has 40% Type-I false-positive rate (Rabosky-Goldberg 2015); only conclude trait-dependent diversification if HiSSE-null is rejected."

What the Agent Will Do

Validate inputs: tree rooted; MSA quality; trait data completeness
Run appropriate framework:
Sequence ASR: PAML codeml + IQ-TREE2 + GRASP for indels
Discrete trait: corHMM with hidden rates + ape::ace for AIC model comparison
Continuous trait: phytools::fastAnc + geiger::fitContinuous for model selection
Bayesian: RevBayes or BayesTraits for full posterior
Apply model selection: ER/SYM/ARD for discrete; BM/OU/EB/lambda for continuous; site-homogeneous vs CAT-PMSF for deep sequence
Stochastic mapping for discrete trait posterior; >=1000 simulations
Quantify confidence: per-site posteriors; phylogenetic signal (Pagel's lambda, Blomberg's K)
Report: ancestral states with uncertainty, model fit, ambiguous sites for downstream testing
Caveats: epistasis, LBA at deep nodes, model adequacy, root sensitivity

Tips

For protein resurrection, use GRASP (handles indels) + design alternative constructs at ambiguous sites (P < 0.8)
corHMM with rate.cat=2 is the modern standard for discrete trait ASR; legacy ape::ace lacks hidden rates
HiSSE is the required null for any trait-diversification correlation test (Rabosky-Goldberg 2015)
For continuous traits, always run BM vs OU vs EB AIC comparison; report Pagel's lambda
Stochastic mapping with pi='fitzjohn' root prior; >=1000 simulations
For deep sequence ASR, use site-heterogeneous models (CAT-PMSF, PhyloBayes) to mitigate LBA
Marginal reconstruction for per-site uncertainty; joint reconstruction for internally consistent ancestor
Multiple-rooting sensitivity tests; STRIDE / MAD rooting alternatives
Document model adequacy: AIC + posterior predictive checks
Codon-level ASR uses PAML M0 (single omega) -- site models are for selection, not ASR
Bayesian methods (RevBayes, MrBayes) provide honest uncertainty; computationally heavy
For convergent rate shifts, RERconverge (categorical phenotype) or CSUBST (amino-acid convergence)
PhyloAcc for noncoding accelerated evolution; not codon-based
Document phylogenetic signal: lambda > 0.7 strong; 0.3-0.7 moderate; < 0.3 weak

comparative-genomics/positive-selection - Branch- and site-level selection on ancestral branches comparative-genomics/ortholog-inference - Orthogroups feed sequence ASR comparative-genomics/gene-tree-species-tree-reconciliation - DTL-aware ancestral gene content comparative-genomics/whole-genome-duplication - Time scale from Ks for trait dating comparative-genomics/comparative-annotation-projection - Project ancestral CDS to descendants phylogenetics/modern-tree-inference - Rooted ML/Bayesian trees phylogenetics/bayesian-inference - RevBayes / MrBayes priors phylogenetics/divergence-dating - Time-calibrated trees for absolute-time ASR alignment/multiple-alignment - PRANK / MACSE indel-aware alignment alignment/alignment-trimming - PREQUAL / HmmCleaner filtering before ASR