Design on What Was Actually Done

PopPIPE transmission analysis and clustering plan — concise

Stepwise experiment plan: (1) sample QC and metadata harmonization; (2) population split by kmer sketching; (3) within-strain split kmer alignments, recombination masking, pairwise SNP/distance matrices; (4) transmission inference using time-aware genetic distance thresholds and phylogenies; (5) cluster validation with epidemiological linkage and sensitivity analyses. Key caveats: recombination obscures clustering signal; transmission trees may not show network clustering when no reinfection occurs

Primary references: PopPIPE description and evaluation (PopPIPE integrates kmer sketching, split kmer analysis and recombination removal) PopPIPE paper and an analytical caveat that clustering may be undetectable in transmission trees without reinfection Limits of detecting network clustering

Visual summary (what to produce first)

Goal

Produce reproducible, auditable transmission inferences across a bacterial population using PopPIPE's two-scale approach: (A) split population into strain-level clusters with fast reference-free kmer sketching, then (B) create high-resolution, recombination-masked alignments inside strains to infer subclusters and potential transmissions. Validate clusters against epidemiological data and test robustness to recombination and analytic choices.

Key conceptual points and limitations

• Reference-free kmer sketching sensitively separates populations by gene content and sequence similarity at broad scales; PopPIPE uses this to define strains before alignment-based subclustering PopPIPE description and repo
• Recombination inflates apparent genetic distances and can create false structure; explicit recombination removal before phylogeny and pairwise distance calculation is essential for accurate transmission inference PopPIPE recombination step
• Network clustering (triangles) in contact networks does not always leave detectable signals in single-outbreak transmission trees, especially without reinfection; interpret topology cautiously and rely on combined genomic+epidemiologic evidence Detectability limits of network clustering

Experimental design and analysis pipeline (detailed, stepwise)

1. Input data and metadata harmonization
   • Required inputs: cleaned Illumina reads or assemblies; sample collection date (at least day precision), location (ward/facility), patient/staff pseudonym, specimen type.
   • QC: fastp/fastqc filtering, remove low coverage samples (<20x for outbreak resolution), check contamination with Kraken2 and remove mixed samples.
   • Metadata standards: ISO8601 dates, standardized location codes, and a column describing epidemiological links (room, admission, exposure events).
2. Run PopPIPE initial population split
   • Use PopPIPE's kmer sketching step (PopPUNK or internal sketching) to partition the dataset into strain groups; this step is fast and reference-free and improves downstream alignment sensitivity PopPIPE kmer partitioning
   • Produce UMAP or neighbor-network visual of sketch distances colored by date/location to spot lab/metadata issues.
3. Within-strain high-resolution analysis
   • For each strain cluster: generate split kmer alignments (or high-quality reference mapping if an appropriate closed reference exists), use recombination detection (Gubbins or fastGEAR or PopPIPE built-in) to mask recombined regions, and create a recombination-masked alignment for phylogeny and pairwise SNP calling.
   • Compute pairwise SNP distances and pairwise date differences; annotate pairwise distances with masked fraction and mapping coverage.
4. Transmission candidate detection rules
   • Define initial genetic thresholds empirically for each organism/strain using background pairwise distance distributions and published mutation rate estimates; e.g., for many bacteria plausible direct transmission often corresponds to 0–10 core SNPs in short timeframes but thresholds vary by species and recombination rate — derive thresholds with sensitivity analysis rather than fixed arbitrary cutoffs Thresholds are dataset specific
   • Combine genetic window with temporal plausibility: require plausible infection-to-infection interval (use known generation interval or sampling interval) and epidemiological plausibility (shared ward, known contact). Score candidate pairs by combined evidence (genetic distance percentile, date gap zscore, co-location binary) and rank.
5. Phylogenetic and transmission inference
   • For subclusters with multiple candidate pairs, infer time-scaled phylogeny (treedater/TreeTime/BEAST for small clusters) on recombination-masked alignment and map sample dates; interpret short branches and topology along with patient movements for likely directionality.
   • Where appropriate, use outbreak transmission inference tools that account for unsampled intermediates and within-host diversity (TransPhylo, SCOTTI), but interpret results with caution because recombination and sampling bias affect inference.
6. Cluster validation and sensitivity analyses
   • Validate clusters by: (a) cross-checking with epidemiological links, (b) bootstrapping phylogenies and re-running clustering thresholds, (c) varying recombination masking stringency, and (d) testing alternative distance metrics (kmer Jaccard, core SNPs).
   • Test detectability limits: simulate outbreaks on contact networks with varying clustering and run PopPIPE-like analysis to estimate false negative and false positive cluster detection rates (informed by the finding that network clustering may be invisible in trees without reinfection) Simulation limits on clustering detection
7. Reporting and visualization
   • Create interactive dashboards for each strain: strain sketch map, within-strain SNP histogram with threshold overlays, recombination-masked phylogeny with metadata tracks, and a ranked transmission candidate table with links to raw reads and alignment QC metrics (PopPIPE includes interactive visualizations out of the box) PopPIPE outputs and visuals

Recommended tools, libraries and reproducible workflow

• PopPIPE repository and docker image (primary pipeline) PopPIPE repo and docker
• Workflow manager: Snakemake or Nextflow for reproducibility and easy reconfiguration.
• QC and preprocessing: fastp, Kraken2, or similar.
• Sketching and clustering: PopPUNK / Mash / sourmash (PopPIPE uses kmer sketching internally).
• Split kmer and alignment: SKA or split-kmer tools as used in PopPIPE; recombination: Gubbins or fastGEAR depending on genome characteristics.
• Phylogenetics/time-scaling: IQ-TREE or RAxML for ML trees; TreeTime or treedater for rapid time-scaling; BEAST for deep Bayesian inference on small clusters.
• Transmission inference: TransPhylo or SCOTTI for models that include unsampled cases; always accompany with sensitivity analysis.
• Visualization: Plotly for interactive plots, Microreact or Auspice for tree + metadata visualizations.

Statistical design and validation

• Define null models and power analysis via simulation: simulate genomes on contact networks with parameter ranges matching observed mutation and recombination rates; measure true positive and false positive recovery of transmission clusters.
• When deciding distance cutoffs, prefer ROC-style analyses using known epidemiological pairs (if available) and report sensitivity/specificity across thresholds.
• Report uncertainty: bootstrap trees, posterior node date credible intervals, and report how cluster membership changes under parameter perturbations.

Blindspots, biases and limitations

• Sampling bias: incomplete sampling (missing links) can fragment or obscure true transmission chains; results are conditional on sampling frame.
• Recombination: residual unmasked recombination can cause spurious clustering; rigorously report masked fraction and perform sensitivity to masking thresholds.
• Threshold overfitting: fixed SNP cutoffs are arbitrary; use empiric, organism-specific calibration and report uncertainty.
• Network clustering detectability: as shown in simulation studies, contact-network triangle clustering may not leave a strong signature in transmission trees without reinfection, so do not infer contact-network clustering from a single outbreak tree alone Limits on inferring network clustering

Deliverables and outputs

1. Reproducible Snakemake/Nextflow pipeline wrapper that runs PopPIPE docker image and produces: strain sketch maps, within-strain masked alignments, ML and time-scaled trees, pairwise distance matrices, ranked transmission candidate tables, and interactive dashboards.
2. A report describing cluster definitions, thresholds, sensitivity analyses, and a per-cluster summary of epidemiological concordance.
3. Simulations quantifying detection limits (power curves) for cluster recovery given estimated mutation and recombination rates.

Immediate actionable checklist to start

1. Assemble reads/assemblies and metadata into standardized table; run QC and remove problematic samples.
2. Run PopPIPE sketch partitioning to generate strain groups and inspect sketch UMAP/networks for oddities PopPIPE initial sketching
3. Select small target strain clusters for immediate within-strain analysis (to validate parameters) and run split-kmer alignment plus recombination masking; compute pairwise distances and produce SNP histograms.

Quick links and tools

When to run a full AI bioinformatics agent

If you want PopPIPE applied to a specific dataset (reads/assemblies + metadata), or want simulations to quantify detection limits for your organism and sampling frame, click Run AI Biology Analysis to execute an end-to-end analysis and produce interactive plots and tables.

Important evidence sources used for all claims:

• PopPIPE paper and code
• PopPIPE preprint
• Detectability of network clustering in trees