Orchestrates genomic-epidemiology outbreak investigation from pathogen isolates to transmission networks, forking bacterial (snippy -> Gubbins recombination-masking -> IQ-TREE -> TreeTime -> TransPhylo) vs viral (Nextstrain/augur), with parallel MLST typing (cgMLST delegated to epidemiological-genomics/pathogen-typing) and AMR surveillance. Use when committing ONE reference genome for SNP calling (every isolate and distance inherits its coordinates), applying MANDATORY Gubbins recombination-masking on core.full.aln before the tree for recombining bacteria (skipping it inflates the clock 2-5x), gating time-scaling on a temporal-signal test (TempEst R2 >= 0.3), using a pathogen- AND population-specific cluster threshold rather than a universal SNP cutoff, or pinning pangolin-data/Nextclade/Freyja versions for the viral route. Hands mechanism to the epidemiological-genomics component skills; not a re-teach of any single step.
Orchestrates genomic-epidemiology outbreak investigation from pathogen isolates to transmission networks, forking bacterial (snippy -> Gubbins recombination-masking -> IQ-TREE -> TreeTime -> TransPhylo) vs viral (Nextstrain/augur), with parallel MLST typing (cgMLST delegated to epidemiological-genomics/pathogen-typing) and AMR surveillance. Use when committing ONE reference genome for SNP calling (every isolate and distance inherits its coordinates), applying MANDATORY Gubbins recombination-masking on core.full.aln before the tree for recombining bacteria (skipping it inflates the clock 2-5x), gating time-scaling on a temporal-signal test (TempEst R2 >= 0.3), using a pathogen- AND population-specific cluster threshold rather than a universal SNP cutoff, or pinning pangolin-data/Nextclade/Freyja versions for the viral route. Hands mechanism to the epidemiological-genomics component skills; not a re-teach of any single step.
[{"after_typing":"Valid ST assigned (7-locus mlst); cgMLST via chewBBACA is deferred to epidemiological-genomics/pathogen-typing"},{"after_amr":"AMR genes identified; AMRFinderPlus applies its curated per-gene thresholds (it does not use a flat 90% identity cutoff)"},{"after_phylodynamics":"Root-to-tip R2 >=0.3, clock rate plausible"},{"after_transmission":"Transmission pairs consistent with epi data"}]
Nextclade: nextclade dataset list --tag latest sars-cov-2
TB-Profiler: tb-profiler list_db (verify WHO catalogue edition)
If code throws ImportError, AttributeError, or TypeError, introspect the installed
package and adapt the example to match the actual API rather than retrying.
Outbreak Pipeline
"Characterize a pathogen outbreak from my isolate sequences" -> Orchestrate MLST typing, SNP phylogeny, TreeTime time-scaled tree construction, TransPhylo transmission inference, AMR profiling, and variant surveillance for genomic epidemiology.
This is a workflow skill: it owns the chaining decisions and hand-offs, not the internals of any one step.
The governing principle
A transmission tree is the MAP estimate among many equally probable trees, and its trustworthiness is decided at four seams.
One reference genome for SNP calling, committed before any isolate. snippy calls SNPs against ONE reference; every isolate, the core alignment, and the cluster distances inherit its coordinates. A build/coordinate mismatch (isolates called against different references, or an AMR panel keyed to a different build) fabricates or hides SNPs. Commit the reference (GenBank format for snippy) up front.
Recombination-masking is MANDATORY for recombining bacteria, on core.full.aln, before the tree. Gubbins masking is not optional for S. pneumoniae, N. gonorrhoeae, E. coli, Klebsiella, Campylobacter, H. pylori — skipping it inflates the clock rate 2-5x and biases R_e. Input MUST be core.full.aln (full positions incl. invariant), NOT core.aln (variable-only). Skip masking ONLY for documented-clonal Mtb.
Time-scaling requires a passing temporal-signal test, committed before trusting any dated tree. TempEst root-to-tip regression (Rambaut 2016) with R2 >= 0.3 as a field convention (the paper sets no threshold and cautions R2 is not a valid significance test); the date-randomisation test is a secondary check (can pass with narrow sampling windows), NOT a substitute. The outbreak-scale clock is lineage/host-specific, not a universal constant.
The cluster threshold is pathogen- AND population-specific — never a universal cutoff. From the literature per pathogen (Mtb <=12/<=5 SNP; S. aureus <=15; K. pneumoniae KPC <=21; C. difficile <=2 masked; Salmonella <=5 cgMLST alleles). Walker's 5-SNP Mtb threshold was calibrated in low-transmission UK and inflates apparent recent transmission 2-5x in high-burden settings. A genomic distance is not an epidemiological distance without a time-scaled prior.
Made-once commitments
Commitment
Consequence inherited downstream
Reference genome + build (bacterial)
Every isolate's SNPs, the core alignment, cluster distances; a mismatch fabricates/hides SNPs
Recombination-masking scheme (Gubbins on core.full.aln)
The clock rate and R_e; skipping inflates the clock 2-5x for recombining taxa
The lineage call; same genome, different call across versions — pin them
Workflow Overview
Pathogen Isolate Genomes (FASTA/FASTQ) + collection dates + (optional) contact data
|
v
+---------+---------+
| |
v v
[1a. MLST + serotyping [1b. AMR + species mode:
+ Pangolin/UShER AMRFinderPlus --organism,
for SARS-CoV-2; TB-Profiler for Mtb,
cgMLST -> hAMRonization across tools]
pathogen-typing]
| |
+--------+----------+
|
v
[2. snippy + snippy-core (bacteria) -> Gubbins on core.full.aln to mask recombination
(mandatory for bacteria; skip only for clonal Mtb)]
|
v
[3. IQ-TREE on recombination-masked alignment + TempEst R^2 >= 0.3 + date-randomisation;
TreeTime --coalescent skyline --clock-filter 4 OR BactDating;
BEAST2 BDSKY (origin > rootHeight, multi-chain) for posterior R_e]
|
v
[4. Transmission inference: outbreaker2 (dense + contact data) OR TransPhylo (sparse,
from dated tree) OR transcluster (pair-level probability); pathogen-specific SNP
threshold for cluster definition -- NEVER a universal cutoff]
|
v
Transmission tree posterior + R_e(t) + lineage / clone context + AMR phenotype
Goal: Assign 7-locus PubMLST sequence types to all isolates for clonal-context interpretation.
Approach: Run Seemann's mlst per assembly; auto-detect scheme; concatenate the per-isolate output into a cohort TSV.
#!/bin/bash
ISOLATES="isolate1.fasta isolate2.fasta isolate3.fasta"
OUTDIR="outbreak_results"mkdir -p ${OUTDIR}/{mlst,amr,alignment,phylo,transmission}
# Run MLST on all isolatesecho"=== MLST Typing ==="for fasta in$ISOLATES; do
sample=$(basename$fasta .fasta)
mlst $fasta > ${OUTDIR}/mlst/${sample}.mlst.txt
done# Combine resultscat${OUTDIR}/mlst/*.mlst.txt > ${OUTDIR}/mlst/all_mlst.tsv
echo"MLST complete: ${OUTDIR}/mlst/all_mlst.tsv"
Step 1b: AMR Detection (Parallel) -- AMRFinderPlus with species mode
Goal: Produce per-isolate AMR calls with species-specific point-mutation panel activated, then harmonise across the cohort to the PHA4GE schema for cross-lab comparison.
Approach: AMRFinderPlus -n for nucleotide assembly with --organism and --plus; pipe each per-isolate TSV through hamronize amrfinderplus with mandatory PHA4GE metadata; hamronize summarize merges to a cohort table. For M. tuberculosis, switch to TB-Profiler -- AMRFinderPlus has no Mtb organism mode.
For M. tuberculosis, route to TB-Profiler instead -- AMRFinderPlus has no Mtb organism mode. For colistin / mcr surveillance and any plasmid-mobility claim, follow with MOB-suite (mob_recon + mob_typer) to determine plasmid context. See epidemiological-genomics/amr-surveillance for the full decision tree.
Goal: Build a recombination-aware core-genome alignment that is safe for downstream clock inference.
Approach: Snippy per isolate against the reference; snippy-core to merge into the core alignment; Gubbins on core.full.aln (NOT core.aln) to mask recombinant tracts. Skipping recombination masking inflates the clock rate 2-5x for recombining bacteria (S. pneumoniae, N. gonorrhoeae, E. coli, Klebsiella, Campylobacter, H. pylori); the date-randomisation test is NOT a sufficient guard.
echo"=== Core Genome Alignment ==="
REFERENCE="reference.gbk"# Reference genome in GenBank format# Run snippy for each isolatefor fasta in$ISOLATES; do
sample=$(basename$fasta .fasta)
snippy --outdir ${OUTDIR}/alignment/snippy_${sample} \
--ref $REFERENCE \
--ctgs $fasta \
--cpus 8
done# Core SNP alignment
snippy-core --ref $REFERENCE --prefix core ${OUTDIR}/alignment/snippy_*
# Mandatory for recombining bacteria (S. pneumoniae, N. gonorrhoeae, E. coli, Klebsiella,# Campylobacter, H. pylori). Skip ONLY for clonal Mtb cross-lineage analyses where# recombination is documented to be rare; even then a recombination check is defensible.# Input MUST be core.full.aln (full positions including invariant); core.aln (variable-only)# gives wrong recombination calls because Gubbins cannot estimate background SNP density.
run_gubbins.py --prefix gubbins core.full.aln
mv core.* gubbins.* ${OUTDIR}/alignment/
echo"Recombination-masked alignment: ${OUTDIR}/alignment/gubbins.filtered_polymorphic_sites.fasta"
Step 3: Phylodynamics with TreeTime
Goal: Time-scale the recombination-masked phylogeny with a global clock-rate estimate, gated by temporal-signal QC.
Approach: IQ-TREE on the recombination-masked alignment with +ASC ascertainment correction; TreeTime with coalescent skyline prior and --clock-filter 4; inspect root_to_tip_regression.pdf BEFORE trusting downstream output (R^2 >= 0.3 minimum as a field convention; TempEst sets no threshold).
import subprocess
from Bio import Phylo, AlignIO
import pandas as pd
import matplotlib.pyplot as plt
from pathlib import Path
outdir = Path('outbreak_results')
# Build ML tree on the recombination-masked alignment with ascertainment-bias correction# +ASC is required because the input contains variable positions only post-Gubbins
subprocess.run([
'iqtree2', '-s', str(outdir / 'alignment/gubbins.filtered_polymorphic_sites.fasta'),
'-m', 'GTR+G+ASC', '-B', '1000', '-bnni', '-T', 'AUTO',
'--prefix', str(outdir / 'phylo/outbreak')
], check=True)
# Prepare metadata with dates# Format: name\tdate (YYYY-MM-DD or decimal year)
metadata = pd.DataFrame({
'name': ['isolate1', 'isolate2', 'isolate3', 'isolate4', 'isolate5'],
'date': ['2024-01-15', '2024-01-22', '2024-02-01', '2024-02-10', '2024-02-15']
})
metadata.to_csv(outdir / 'phylo/metadata.tsv', sep='\t', index=False)
# Run TreeTime
subprocess.run([
'treetime',
'--tree', str(outdir / 'phylo/outbreak.treefile'),
'--aln', str(outdir / 'alignment/gubbins.filtered_polymorphic_sites.fasta'),
'--dates', str(outdir / 'phylo/metadata.tsv'),
'--outdir', str(outdir / 'phylo/treetime_output'),
'--coalescent', 'skyline',
'--clock-filter', '4', # SD multiplier for TreeTime's clock filter'--confidence',
'--reroot', 'best'
], check=True)
# Temporal-signal QC: inspect root_to_tip_regression.pdf BEFORE trusting any downstream output.# R^2 >= 0.3 minimum (field convention, NOT from Rambaut 2016 -- TempEst sets no threshold and# states R^2 is an informal dispersion measure, not a significance test). If R^2 < 0.3, time-scaling is not# supported -- report uncertainty and consider extending the sampling window. The# date-randomisation test is a secondary check; it can pass with narrow sampling windows# (false negative).print('TreeTime output:', outdir / 'phylo/treetime_output')
Step 4: Transmission Inference with TransPhylo
Goal: Reconstruct the posterior who-infected-whom transmission tree and R_e from the dated phylogeny.
Approach: Convert the TreeTime dated tree to TransPhylo ptree; supply pathogen-tuned generation-time and sampling-time Gamma priors; run MCMC at >=1e5 iterations (10k is smoke-test only); summarise via medoid transmission tree and per-pair WIWS probability. For dense outbreaks with contact-tracing data, outbreaker2 with ctd is preferred over TransPhylo (genomic-only).
library(TransPhylo)
library(ape)# Load dated tree from TreeTime
tree <- read.nexus("outbreak_results/phylo/treetime_output/timetree.nexus")# Set parameters# dateT: date when sampling stopped# w.shape, w.scale: generation time distribution (Gamma)# For many bacteria: mean ~14 days, shape=2, scale=7
dateT <- 2024.2 # Decimal year when sampling ended (end of observation)
w_shape <- 2 # Generation time shape (Gamma)
w_scale <- 7/365# Gamma SCALE = 7 days; mean generation time = shape*scale = 2*7 = ~14 days# TransPhylo operates on a `ptree` (dated phylogeny + last-sample date), NOT a raw ape phylo;# convert first or inferTTree errors on a NULL ptree$ptree/$nam.
ptree <- ptreeFromPhylo(tree, dateLastSample = dateT)# Run TransPhylo with enough iterations for posterior convergence; 10k is a smoke-test only.# For publication, run >=1e5 (small outbreaks) to >=1e6+ iterations and inspect trace plots.
res <- inferTTree(ptree, dateT = dateT,
w.shape = w_shape, w.scale = w_scale,
mcmcIterations =1e5,
startNeg =1, startPi =0.5)# medTTree returns a coloured transmission tree (ctree); plot it with plotCTree
med_ctree <- medTTree(res)# Plot transmission tree
pdf("outbreak_results/transmission/transmission_tree.pdf", width=10, height=8)
plotCTree(med_ctree)
dev.off()# Who infected whom matrix (same 0.5 burn-in as the R_e estimate below, so both discard pre-convergence)
wiw <- computeMatWIW(res, burnin =0.5)
write.csv(wiw,"outbreak_results/transmission/who_infected_whom.csv")# R_e estimate (effective reproduction number under current immunity / interventions).# This is NOT R_0 (basic reproduction number in a fully susceptible population);# the phylodynamics literature is explicit about this distinction.# getOffspringDist(record, burnin, k) gives the per-case offspring distribution; average# across sampled hosts for a cohort R_e (or use BEAST2 BDSKY for a posterior Re(t)).# Host names come from the ptree (res has no $ttree$nam field).
offspring <- sapply(ptree$nam,function(k) mean(getOffspringDist(res, k = k, burnin =0.5)))# The interval is the 2.5-97.5% spread of per-host mean offspring (across-host dispersion), NOT a# posterior credible interval (per-host posteriors were collapsed by mean() first).
cat("R_e estimate:", mean(offspring),"(across-host 2.5-97.5%:", quantile(offspring,0.025),"-", quantile(offspring,0.975),")\n")
Python Alternative: TransPhylo via rpy2
Goal: Drive the same TransPhylo workflow from Python pipelines that prefer not to fork into R.
Approach: rpy2 bridges into the TransPhylo R package with named-argument passing; same priors and MCMC iteration discipline apply.
import rpy2.robjects as ro
from rpy2.robjects.packages import importr
from rpy2.robjects import pandas2ri
import pandas as pd
from pathlib import Path
pandas2ri.activate()
transphylo = importr('TransPhylo')
ape = importr('ape')
outdir = Path('outbreak_results')
tree = ape.read_nexus(str(outdir / 'phylo/treetime_output/timetree.nexus'))
date_t = 2024.2
w_shape = 2
w_scale = 7/365# Convert to a TransPhylo ptree before inference (inferTTree needs ptree, not a raw phylo).
ptree = transphylo.ptreeFromPhylo(tree, dateLastSample=date_t)
res = transphylo.inferTTree(ptree, dateT=date_t, w_shape=w_shape, w_scale=w_scale,
mcmcIterations=10000, startNeg=1, startPi=0.5)
# medTTree returns a ctree; hand it to R's global env and plot with plotCTree.
med_ctree = transphylo.medTTree(res)
ro.globalenv['med_ctree'] = med_ctree
ro.r(f'''
pdf("{outdir}/transmission/transmission_tree.pdf", width=10, height=8)
plotCTree(med_ctree)
dev.off()
''')
print(f'Transmission tree saved to {outdir}/transmission/')
Visualization: Outbreak Timeline
Goal: Plot isolates over time coloured by sequence type to communicate cluster expansion and clonal context.
Approach: Merge collection-date metadata with MLST output; plot per-isolate timestamps as a strip chart with per-ST colour.
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from datetime import datetime
metadata = pd.read_csv('outbreak_results/phylo/metadata.tsv', sep='\t')
metadata['date'] = pd.to_datetime(metadata['date'])
mlst = pd.read_csv('outbreak_results/mlst/all_mlst.tsv', sep='\t', header=None,
names=['file', 'scheme', 'ST'] + [f'locus{i}'for i inrange(7)])
mlst['sample'] = mlst['file'].apply(lambda x: x.split('/')[-1].replace('.fasta', ''))
# Merge data
combined = metadata.merge(mlst[['sample', 'ST']], left_on='name', right_on='sample')
fig, ax = plt.subplots(figsize=(12, 6))
colors = {'ST11': 'red', 'ST258': 'blue', 'ST307': 'green'}
for st in combined['ST'].unique():
subset = combined[combined['ST'] == st]
ax.scatter(subset['date'], [1]*len(subset), label=f'ST{st}',
s=100, c=colors.get(f'ST{st}', 'gray'), alpha=0.7)
ax.set_xlabel('Date')
ax.set_ylabel('')
ax.set_title('Outbreak Timeline by Sequence Type')
ax.legend()
ax.xaxis.set_major_formatter(mdates.DateFormatter('%Y-%m-%d'))
plt.xticks(rotation=45)
plt.tight_layout()
plt.savefig('outbreak_results/outbreak_timeline.pdf')
Parameter Recommendations
Step
Parameter
Value
Rationale
snippy
--mincov
10
Minimum coverage for variant call
Gubbins
input
core.full.aln
Full positions required to estimate background SNP density; core.aln is wrong
IQ-TREE
-m
GTR+G+ASC
+ASC ascertainment correction for SNP-only post-Gubbins input
TreeTime
--clock-filter
4
SD multiplier on root-to-tip residual; TreeTime convention
TreeTime
R^2 minimum
0.3
Below this, temporal signal treated as insufficient (field convention; TempEst itself sets no cutoff)
TransPhylo
w.shape, w.scale
2, 7/365
Gamma scale 7 days x shape 2 = ~14-day mean generation time; cite the pathogen-specific literature
TransPhylo
mcmcIterations
1e5-1e6+
10k is a smoke-test only; inspect trace and ESS before reporting
Single-chain ESS >=200 is necessary but not sufficient; combine after marginal overlap
Pangolin
--analysis-mode
usher
pangoLEARN deprecated mid-2023; UShER default since v4 (de Bernardi Schneider 2024, Virus Evol 10:vead085)
Pathogen-Specific SNP / cgMLST Cluster Thresholds
Cluster definition is pathogen- AND population-specific. NEVER apply a universal cutoff. See epidemiological-genomics/transmission-inference for full table with citations.
epidemiological-genomics/variant-surveillance - Nextclade for viral outbreaks
phylogenetics/modern-tree-inference - IQ-TREE2 model selection
References
Croucher NJ, Page AJ, Connor TR, et al (2015) Rapid phylogenetic analysis of large samples of recombinant bacterial whole genome sequences using Gubbins. Nucleic Acids Research 43:e15. DOI 10.1093/nar/gku1196. (recombination masking.)
Didelot X, Fraser C, Gardy J, Colijn C (2017) Genomic infectious disease epidemiology in partially sampled and ongoing outbreaks (TransPhylo). Molecular Biology and Evolution 34:997-1007. DOI 10.1093/molbev/msw275. (transmission inference.)
Walker TM, Ip CLC, Harrell RH, et al (2013) Whole-genome sequencing to delineate Mycobacterium tuberculosis outbreaks: a retrospective observational study. Lancet Infectious Diseases 13:137-146. DOI 10.1016/S1473-3099(12)70277-3. (5-SNP threshold, low-transmission calibration.)
Sagulenko P, Puller V, Neher RA (2018) TreeTime: maximum-likelihood phylodynamic analysis. Virus Evolution 4:vex042. DOI 10.1093/ve/vex042.
