Key Points

Background

Captivity can alter diet, exposure and behaviour — all of which may reshape the gut microbiome.
The dataset contains a small, balanced subset (5 captive, 5 wild) suitable for teaching and demonstrating methods.
Use byobu-screen to keep long‑running commands alive across disconnections.
Symlinks point to /mnt/shared_data to conserve storage and keep everyone working from the same files.

Confirm input format (CasavaOneEightSingleLanePerSampleDirFmt) when importing.
Primer removal is essential; untrimmed primers can disrupt denoising and downstream inference.
Inspect interactive quality plots and consider read overlap (~50 bp recommended) when setting truncation lengths for paired‑end merging.
Use dada2 denoise‑paired with appropriate pooling (here ‘pseudo’) and monitor read retention in denoising_stats outputs.

Classifier must be trained for the same primer/region used in the dataset; mismatched classifiers give poor results.
The classify‑sklearn step can be memory intensive — precomputed classifications are provided for the workshop.
Retain unassigned ASVs (they may still be biologically meaningful), but exclude obvious host/plant contaminants.
Low‑abundance filtering is generally unnecessary with the ASV paradigm unless specific reasons exist.

Phylogenetic distances are required for phylogenetic alpha and beta metrics (Faith’s PD, UniFrac).
Use align‑to‑tree‑mafft‑fasttree for a reproducible single‑threaded workflow.
Keep track of the unrooted and rooted tree artefacts for downstream use.

Explore taxonomic composition at multiple taxonomic levels and by metadata categories (e.g. Captivity).
Choose sampling/depth parameters informed by the feature‑table summary and rarefaction curves to balance sample retention and depth.
Unweighted and weighted UniFrac capture different aspects (presence/absence vs abundance) of community dissimilarity.
ANCOM‑BC2 is compositionally aware and suitable for differential abundance while applying bias correction.

Exported artefacts include: tree (Newick), feature‑table.biom, taxonomy.tsv and representative sequences if required.
Use biom convert to produce TSV tables and sed (or equivalent) to remove extra header lines.
Confirm consistent ASV order between taxonomy and table files before importing into phyloseq or other R packages.

Training a classifier requires region‑matched reference reads and taxonomy; forward/reverse primers must be provided when extracting reads.
Classifier training and full classify‑sklearn runs can be computationally demanding — Consider precomputed classifiers or higher memory/CPU resources for large datasets.