Bio Ecological Genomics Species Delimitation

Species Delimitation

Delimits putative species from molecular data using complementary distance-based, tree-based, and coalescent methods.

Overview of Methods

ASAP (Assemble Species by Automatic Partitioning)

Distance-based method that finds the optimal partition of sequences into putative species:

# ASAP web interface: https://bioinfo.mnhn.fr/abi/public/asap/asapweb.html
# Upload aligned FASTA, select distance model

# Command-line ASAP (if available)
ASAP -i aligned_sequences.fasta -d K2P -o asap_results/
# -d K2P: Kimura 2-parameter distance (standard for COI barcoding)
# -d p: uncorrected p-distance (for closely related taxa)
# -d JC: Jukes-Cantor (simplest substitution model)

Interpreting ASAP Results

# ASAP ranks partitions by asap-score (lower = better)
# Reports: partition rank, number of groups, asap-score, p-value, threshold distance
# Always examine top 5-10 partitions, not just the best one
# Large gaps between consecutive asap-scores indicate robust partitioning

Python ASAP-Style Analysis

from Bio import AlignIO, Phylo
from Bio.Phylo.TreeConstruction import DistanceCalculator, DistanceTreeConstructor
from scipy.cluster.hierarchy import fcluster, linkage
from scipy.spatial.distance import squareform
import numpy as np

alignment = AlignIO.read('aligned_sequences.fasta', 'fasta')

calculator = DistanceCalculator('identity')
dm = calculator.get_distance(alignment)

names = [r.id for r in alignment]
n = len(names)
dist_array = np.zeros((n, n))
for i in range(n):
    for j in range(n):
        dist_array[i][j] = dm[names[i], names[j]]

condensed = squareform(dist_array)
Z = linkage(condensed, method='average')

# Scan thresholds to identify barcode gap
# COI barcode gap typically at 2-3% for animals
thresholds = np.arange(0.01, 0.10, 0.005)
for t in thresholds:
    clusters = fcluster(Z, t=t, criterion='distance')
    n_clusters = len(set(clusters))
    print(f'Threshold {t:.3f}: {n_clusters} groups')

bPTP (Bayesian Poisson Tree Processes)

Tree-based method that models speciation as a branching rate shift:

# iTaxoTools PTP (Python 3 compatible)
pip install PTP-pyqt5

# Run bPTP from command line
python -m PTP.PTP -t rooted_tree.nwk -o bptp_results \
    --type bayesian --ngen 100000 --burnin 0.1 --seed 42

# --type bayesian: Bayesian MCMC (preferred over ML for support values)
# --ngen 100000: MCMC generations; increase for large trees
# --burnin 0.1: discard first 10% of samples

Interpreting bPTP Output

# Support values for each species partition:
#   > 0.95: strong support
#   0.80-0.95: moderate support
#   < 0.80: weak support, boundary uncertain
# bPTP can over-split when populations have strong geographic structure
# Compare against distance-based methods to identify over-splitting

GMYC (Generalized Mixed Yule Coalescent)

Models the transition between interspecific (Yule) and intraspecific (coalescent) branching rates on an ultrametric tree:

library(splits)
library(ape)

# Step 1: Prepare ultrametric tree
# GMYC requires an ultrametric (time-calibrated) tree
tree <- read.tree('rooted_tree.nwk')

# chronos() for quick ultrametric conversion (penalized likelihood)
# lambda=1: smoothing parameter; higher = more clock-like
ultrametric_tree <- chronos(tree, lambda = 1)
class(ultrametric_tree) <- 'phylo'

# Verify ultrametric property
is.ultrametric(ultrametric_tree)

# Step 2: Run single-threshold GMYC
gmyc_result <- gmyc(ultrametric_tree, method = 'single')

# Summary: number of entities, confidence interval, likelihood ratio test
summary(gmyc_result)

# Number of putative species (ML entities)
cat('GMYC species:', gmyc_result$entity[1], '\n')

# Likelihood ratio test p-value
# p < 0.05: significant speciation-coalescent transition detected
cat('LR test p-value:', gmyc_result$p.value[1], '\n')

# Step 3: Extract species assignments
species_list <- spec.list(gmyc_result)
cat('Species assignments:\n')
print(species_list)

Multiple-Threshold GMYC

# Allows different coalescent-to-speciation transitions across the tree
# Better for datasets with variable population sizes or divergence times
gmyc_multi <- gmyc(ultrametric_tree, method = 'multiple')
summary(gmyc_multi)

GMYC Visualization

# Color-coded tree by species partition
pdf('gmyc_species_tree.pdf', width = 10, height = 12)
plot(gmyc_result, cex = 0.6)
dev.off()

# Support surface: likelihood across threshold positions
pdf('gmyc_support.pdf', width = 8, height = 5)
plot(gmyc_result, type = 'support')
dev.off()

BPP (Bayesian Phylogenetics and Phylogeography)

Full multispecies coalescent model for rigorous species delimitation:

# BPP control file for species delimitation (A10 analysis)
# A10: joint species delimitation and species tree estimation

seed = 12345

seqfile = alignment.phy
Imapfile = imap.txt      # maps individuals to putative species
outfile = bpp_results.txt
mcmcfile = bpp_mcmc.txt

speciesdelimitation = 1   # 1 = delimitation mode
speciestree = 1           # 1 = estimate species tree
speciesmodelprior = 1     # 1 = uniform prior on all rooted trees

# species&tree: species names, sample sizes, starting tree
species&tree = 4  speciesA speciesB speciesC speciesD
                  10 8 12 6
                  ((speciesA, speciesB), (speciesC, speciesD));

# Prior on theta (ancestral population size)
# thetaprior = alpha beta: inverse-gamma(alpha, beta)
# alpha=3, beta=0.004: mean=0.002 (typical for moderate diversity)
thetaprior = 3 0.004

# Prior on tau (divergence time)
# tauprior = alpha beta: inverse-gamma(alpha, beta)
# alpha=3, beta=0.002: mean=0.001 (relatively recent divergence)
tauprior = 3 0.002

# MCMC settings
nsample = 100000
sampfreq = 2
burnin = 50000

Interpreting BPP Results

# Posterior probability for each delimitation model
# PP > 0.95: strong support for that delimitation
# PP 0.50-0.95: moderate support
# PP < 0.50: uncertain, consider lumping species
# Always run multiple independent chains to check convergence

Multi-Method Comparison

Compare delimitation results across methods to identify consensus species:

library(ape)

# Partition vectors from each method (species assignment per individual)
asap_partition <- c(1, 1, 1, 2, 2, 3, 3, 3, 4, 4)
bptp_partition <- c(1, 1, 1, 2, 2, 3, 3, 4, 5, 5)
gmyc_partition <- c(1, 1, 1, 2, 2, 3, 3, 3, 4, 4)

# Compare with adjusted Rand index
# ARI = 1: perfect agreement; ARI = 0: random; ARI < 0: less than random
library(fossil)
cat('ASAP vs bPTP ARI:', adj.rand.index(asap_partition, bptp_partition), '\n')
cat('ASAP vs GMYC ARI:', adj.rand.index(asap_partition, gmyc_partition), '\n')
cat('bPTP vs GMYC ARI:', adj.rand.index(bptp_partition, gmyc_partition), '\n')

# Consensus: individuals assigned to the same species by >= 2 methods
comparison <- data.frame(
    individual = paste0('ind_', 1:10),
    ASAP = asap_partition,
    bPTP = bptp_partition,
    GMYC = gmyc_partition
)

# Count number of species per method
cat('\nSpecies counts:\n')
cat('  ASAP:', length(unique(asap_partition)), '\n')
cat('  bPTP:', length(unique(bptp_partition)), '\n')
cat('  GMYC:', length(unique(gmyc_partition)), '\n')

Ultrametric Tree Preparation

Most tree-based methods require ultrametric input:

library(ape)
library(phytools)

tree <- read.tree('ml_tree.nwk')

# Method 1: chronos (penalized likelihood, fastest)
# model='correlated': rates correlated across branches (relaxed clock)
ultra_tree <- chronos(tree, model = 'correlated')

# Method 2: BEAST2 (Bayesian, most rigorous but slow)
# Produces ultrametric tree with credibility intervals
# Run externally, import the MCC tree

# Verify
is.ultrametric(ultra_tree)

# Fix near-ultrametric trees with small rounding errors
ultra_tree <- force.ultrametric(ultra_tree, method = 'extend')

Related Skills

• edna-metabarcoding - Generate barcode sequences from eDNA
• conservation-genetics - Population-level genetic assessment
• phylogenetics/tree-io - Tree input/output for delimitation methods
• phylogenetics/modern-tree-inference - Phylogenetic tree construction
• database-access/entrez-fetch - Retrieve barcode sequences from GenBank