Computes assembly metrics and taxonomic lineage for prokaryotic or eukaryotic input. Prokaryotes: QUAST for assembly stats, CheckM2 for completeness/contamination, GTDB-Tk (classify_wf) for taxonomy. Eukaryotes: QUAST, EukCC for completeness/redundancy, CAT (via MetaEuk-predicted proteins) against NCBI taxonomy. A unified High-Quality (HQ) filter — completeness ≥ 90%, contamination ≤ 5%, taxonomic resolution to at least Genus — gates what proceeds downstream; eukaryotic assemblies stop here and aren't propagated further.

Prokaryotic path (is_euk: false)

• QUAST — contig count, total length, GC%, N50/L50 assembly statistics
• CheckM2 — machine-learning predict workflow using the uniref100 database; accounts for lineage-specific marker gene presence and copy number
• GTDB-Tk classify_wf — identifies 120 bacterial + 15 archaeal marker genes to place genomes within the GTDB reference tree; confirms species-level ID via ANI against type strains
• Prokaryotic_genome_summary.py — parses GTDB classification into discrete taxonomic ranks and merges all outputs

Eukaryotic path (is_euk: true)

• QUAST — same assembly metrics as the prokaryotic path
• EukCC — single-genome mode with lineage-specific marker sets for completeness and redundancy estimation
• CAT — uses protein sequences predicted during the EukCC run (via the MetaEuk engine) for protein-level NCBI taxonomy classification; fragment bit-score cutoff (-f) 0.5, r-value (-r) 3
• Eukaryotic_genome_summary.py — merges outputs via outer joins on assembly identifiers

High-Quality (HQ) filter thresholds

Completeness   >= 90%    (CheckM2 or EukCC)
Contamination  <=  5%    (prokaryotes only)
Taxonomy: Domain through Genus must be fully resolved

Passes -> genome_summary.csv            -> feeds Module 2
Fails  -> unqualified_genome_summary.csv -> excluded

Determines taxonomic position and genomic distance against validly published type strains, dynamically retrieved by genus. Computes three relatedness metrics in parallel — ANI (FastANI), AAI (aai.rb over Prokka-predicted proteomes), and POCP (Qin et al. 2014 method) — plus an optional 16S route via Barrnap + BLASTn. Genomes with WGS ANI < 95% are flagged as potential novel species and isolated for downstream phylogenomic validation.

16S rRNA workflow (is_16S: true)

• Input FASTA files with 16S sequences are queried against the NCBI 16S Ribosomal RNA Database using BLASTn
• compile_results_16S.py — filters hits to the highest % identity per query, fetches NR/NG accessions and official organism names via Biopython Entrez
• Output: compiled_results.csv mapping each strain to its closest validly published relative with % identity and accession

WGS pipeline — 4 phases

• Phase 1 — 16S extraction: Barrnap predicts rRNA genes → BLASTn against the NCBI 16S DB identifies the host genus
• Phase 2 — Reference acquisition: NCBI TaxID retrieved → Datasets API queries all RefSeq (GCF) assemblies for that genus → download_genomes.sh bulk-downloads with verification and indexing
• Phase 3 — ANI: FastANI compares each query against the full genus reference set; the best-matched reference (highest ANI) is identified for proteome-based steps
• Phase 4 — AAI + POCP: aai.rb computes two-way AAI using reciprocal BLASTp (min 50 hits, 20% identity); pocp.sh calculates POCP per Qin et al. 2014 (40% aa identity, 50% coverage, E-value < 1×10⁻⁵)

Three genomic distance indices

• ANI — Average Nucleotide Identity: species-level boundary at 95%. Below = novel.
• AAI — Average Amino Acid Identity: assesses broader evolutionary distances via reciprocal BLASTp of Prokka-predicted proteomes
• POCP — Percentage of Conserved Proteins: genus-level delineation; < 50% suggests a different genus

compiled_results.csv output columns

Strain #             - query genome
Name (WGS)           - organism name (NCBI Entrez)
Accession # (WGS)    - best GCF reference
ANI (%)              - FastANI identity
AAI (%)              - two-way amino acid identity
PoCP%                - % conserved proteins
Name (16S)           - 16S rRNA best hit
Accession # (16S)    - NR accession
% similarities (16S) - BLASTn identity

Automatic outgroup & HMM-set selection for Module 3

As the last step of its own run, Module 2 resolves an outgroup genome and HMM set for every genus among the novel (ANI < 95%) candidates, writing them straight into genome_summary_for_tree.csv — Module 3 needs nothing filled in by hand for a normal run:

• HMM set — selected deterministically from the genome's GTDB phylum and class, no external lookup needed.
• Outgroup genome — resolved by trying progressively broader taxonomic ranks, and at each rank, progressively broader data sources. Starting at the genome's GTDB-Tk Order; if that's unresolved (common for genuinely novel, divergent genomes), moving to Class, then Phylum. At whichever rank is being tried, the GTDB REST API is queried first (authoritative type strains, no key required), falling back to an NCBI Entrez Assembly search at that same rank before widening further.
• A rank/genus combination already resolved in an earlier run is served from a local cache instead of re-querying — unless the cached result is itself empty (a past network failure), which always retries.
• If all three ranks come up empty, the genus is left with no outgroup and Module 3 skips that genus's tree entirely (an unrooted tree isn't useful).

A multi-metric taxonomic decision following established IJSEM standards. Isolates exhibiting ANI below 95% to their closest validly published relative are flagged as potential novel species and staged for downstream phylogenomic validation.

What each index contributes

• ANI (Average Nucleotide Identity) — primary species boundary. < 95% to any type strain = potential novel species
• AAI (Average Amino Acid Identity) — sensitive to broader evolutionary distances; corroborates ANI at the proteome level
• POCP (% Conserved Proteins) — genus-level boundary; < 50% POCP supports classification into a new genus

Output files written

compiled_results.csv   - all genomes (novel + known) with full OGRI metrics
potential_novel.csv    - ANI < 95% candidates only  -> feeds Module 3

prepare_tree_inputs.py auto-stages novel FASTA files
for phylogenomic tree construction

✦ ANI < 95% — candidate novel species: triggers the full M3–M6 characterization pipeline — M3 phylogenetics (GToTree + IQ-TREE confirms an isolated novel branch), M4 metagenomics (sylph profiles natural prevalence), M5 functional characterization, and M6 metabolic modeling.

✓ ANI ≥ 95% — known species: a full evidence table is still generated in compiled_results.csv, combining WGS and 16S identity metrics, organism name, and reference accession.

Places genomes flagged as novel in Module 2 within their taxonomic context. Query genomes are partitioned into genus-specific clusters; conserved single-copy genes are extracted via genus-specific HMMs, and Maximum-Likelihood trees are inferred with bootstrap support. Outputs are formatted for interactive visualization. Genus grouping, HMM set, and outgroup are all filled in automatically by Module 2 — see its card above for how, and how to override it.

Two-stage Snakemake workflow

• gtotree_snakefile — applies genus-specific HMMs via GToTree to extract universal Single-Copy Genes (SCGs); aligns and concatenates SCGs across query genomes, reference type strains, and the outgroup genome
• iqtree_snakefile — infers Maximum Likelihood trees using IQ-TREE on the concatenated SCG alignment; 1000 ultrafast bootstrap replicates for statistical branch support

IQ-TREE run command

iqtree \
  -s Aligned_SCGs.faa \   # concatenated SCG alignment
  -spp Partitions.txt \   # per-gene partition model
  -m MFP \                # ModelFinder Plus
  -bb 1000 \              # ultrafast bootstraps
  -nt 4 \                 # CPU threads
  -pre {genus}_iqtree_out # output prefix per genus

iTOL annotation — tree_annotation.py

• Generates per-genus annotation CSV files formatted for direct iTOL upload
• Colour-coded labels for novel candidate isolates, reference type strains, and the outgroup genome
• Bootstrap support values shown at internal nodes for assessment of novelty placement confidence

Quantifies prevalence and relative abundance of novel species across user-provided metagenomic datasets using k-mer-based containment estimation — fast, and without the computational cost of read alignment. Produces a standardized master abundance table across samples.

Detection thresholds used

containment threshold (c) = 100  ->  species-level resolution (default)
min-number-kmers          = 5    ->  mitigates false positives from
                                     low-complexity sequences

Alternative thresholds:
  c = 95  ->  genus-level
  c = 90  ->  family-level

Workflow steps

• sylph_sketch_db — builds a .syldb sketch database once from all reference genomes (recursive FASTA scan)
• sylph_sketch_sample — sketches metagenomic reads: paired-end (_1/_R1 + _2/_R2 naming convention) via sylph sketch -1/-2, or single-end/unpaired reads via sylph sketch -r if no matching pair is found
• sylph_profile — runs k-mer containment profiling at the set threshold against the reference sketch database
• merge_results.py — concatenates per-sample TSV outputs → final_report.csv with a Sample_ID column added

final_report.csv output columns

• Sample_ID — metagenomic dataset identifier
• Genome — reference genome detected
• Containment — k-mer containment score
• ANI — estimated ANI derived from k-mer similarity
• Reads_Queried — total reads processed per sample
• Reads_Matching — reads matching the reference genome

Multi-modal profiling of genomic features, metabolic potential, resistance mechanisms, and mobile genetic elements: structural annotation and COG classification, secondary metabolite / biosynthetic gene cluster detection, prophage and plasmid identification, and antibiotic/biocide/metal resistance screening against curated databases. Results compile into summary tables for visualization.

Step 1 — Structural annotation

• Prokka identifies protein-coding sequences (CDS), ribosomal RNAs, and transfer RNAs via the Prodigal gene-finding algorithm
• Outputs per genome: .gff (annotation), .faa (protein sequences), .ffn (nucleotide genes), .gbk (GenBank format)

Step 2–3 — COG functional classification

• COGclassifier classifies predicted proteins into Clusters of Orthologous Groups (COGs) against the NCBI Conserved Domain Database (CDD) — 26 functional categories
• cog_merge.py pivots discrete classification counts into a comparative abundance matrix (merged_classifier_count.csv)

Step 4–5 — Secondary metabolite detection

• antiSMASH ("bacteria" taxon setting) detects Biosynthetic Gene Clusters: NRPS, PKS, terpenes, RiPPs, and more
• --cb-general and --cb-knownclusters flags enabled for comparison against the MIBiG database
• parserfile.py — custom HTML parsing engine extracts region coordinates, cluster types, and MIBiG similarity scores → AntiSMASH_results.csv

Step 6 — Resistance, virulence & biocide genes

• ABRICATE screens four curated databases: CARD (Antimicrobial Resistance), VFDB (Virulence Factors), BacMet2 (Biocide & Metal Resistance), NCBI
• High-confidence threshold: ≥ 80% nucleotide identity AND ≥ 80% query coverage enforced across all databases

Step 7 — Mobile element identification

• geNomad "end-to-end" mode predicts prophages and plasmids within each assembly
• genomad_merge.py consolidates individual sample outputs → compiled virus_summary.csv, virus_genes.csv, plasmid_summary.csv, plasmid_genes.csv

Reconstructs a genome-scale metabolic model for each novel species via a top-down template approach, exported in standardized SBML/FBC format. Network topology and biomass connectivity are validated under unconstrained growth tests, and model quality is benchmarked for stoichiometric consistency.

Getting CPLEX (student/academic license)

1. Create an IBMid using your university/academic email address.
2. Go to the IBM Academic Initiative downloads page and sign in.
3. Select IBM ILOG CPLEX Optimization Studio.
4. Search for "CPLEX 22.1.1" specifically — NOSE is verified against this version, not the older ones most generic install guides walk through.
5. Choose the HTTP download option, then check the box for the installer matching your OS (e.g. "...CPLEX Optimization Studio V22.1.1 for Linux x86-64").
6. Accept the license agreement — the Download button only appears once you do.
7. Click Download. The installer file (e.g. cplex_studio2211.linux-x86-64.bin) starts downloading.

Linux / macOS — make the installer executable and run it as superuser, accepting the default install path unless you have a reason to change it:

$ chmod +x cplex_studio2211.linux-x86-64.bin
$ sudo ./cplex_studio2211.linux-x86-64.bin

Windows — run the downloaded .exe installer as Administrator, keeping the default install path.

/opt/ibm/ILOG/CPLEX_Studio2211/cplex/python/3.9/x86-64_linux

CarveMe reconstruction command

carve \
  --dna {genome.fasta} \   # DNA input mode
  --prodigal \             # high-fidelity gene prediction
  -o {sample}.sbml \       # SBML + FBC output
  --solver cplex \         # LP solver
  --diamond-args "-p {threads}"

LP solver options

CPLEX is NOSE's configured solver, as set up above. Gurobi is also supported as an alternative if you'd rather use a Gurobi license instead of CPLEX — set carveme.solver to gurobi in Module 6's config and point Gurobi License Path at your own installation instead of the CPLEX fields. Gurobi offers a free Named-User Academic license for eligible students and staff at gurobi.com/academia — activation (grbgetkey) needs to happen while on your institution's network or VPN.

Model validation — generate_model_stats.py

• COBRApy loads each SBML model and counts reactions, metabolites, and genes with metabolic function
• Unconstrained growth test: all exchange reaction bounds set to ±1000 mmol·gDW⁻¹·h⁻¹; biomass objective defined as the detected biomass reaction
• growth = 0 in the output indicates a disconnected or broken model requiring investigation
• compile_model_summary.py aggregates all individual model TSVs → model_summary.csv for cross-isolate comparison

model_summary.csv columns

sample         - genome name
n_reactions    - total reactions in the GEM
n_metabolites  - total metabolites
n_genes        - genes with metabolic function
growth         - unconstrained max growth rate
               - 0 = model disconnected/broken
memote_score   - MEMOTE quality score (0-100%)

MEMOTE quality benchmarking

• Standardised quality control suite assessing stoichiometric consistency, metabolite formula redundancy, and mass/charge balance
• A MEMOTE snapshot HTML + JSON report is generated per isolate
• Scores cover multiple categories — a high overall score indicates a well-curated, publication-ready GEM
• GEMs exported in SBML/FBC format: compatible with COBRA Toolbox, cobrapy, cameo, OptFlux