SYSTEM OVERVIEW & GUIDE

BranchManager

Marker-Gene Quality Control, Novelty Scoring, & Genome-Candidate Prioritisation Toolkit

Core Mission: Answers the critical question - "Has this microbial lineage already been isolated and characterised, or is it sufficiently novel and underrepresented to justify whole-genome sequencing (WGS) investment?"

1. Executive Summary

BranchManager is an end-to-end biological decision-support platform built for microbiology laboratories, research consortia, and biobanks managing large isolate collections. When processing thousands of physical isolates, determining which candidates warrant high-cost downstream sequencing is a major bottleneck.

BranchManager streamlines this process by evaluating marker-gene sequences (such as 16S rRNA), performing strict quality control on raw instrument chromatograms, classifying sequences across multiple global taxonomy databases, measuring multi-layered biological novelty, and proposing optimal primary and backup candidate sets for whole-genome sequencing.

🔬 Automated Quality Control Evaluates raw Sanger/AB1 chromatogram trace files, trims low-quality ends, detects mixed peaks, quality-weights overlapping reads, and isolates failed sequences for manual review.	🌐 Multi-Lens Novelty Measures sequence divergence independently against internal physical lab baselines (The Hungate Collection), rolling multi-partner candidate collections, and global public references (GTDB, SILVA).	🎯 Balanced Selection Maintains configurable pangenome targets per species, picking primary target candidates to fill genomic gaps while reserving backup strains for extraction failure resilience.

2. Core Operational Ethos & Principles

• Absence of a Hit ≠ Novelty: A sequence lacking close reference hits is not automatically novel. It may represent poor trace quality, chimeric sequences, contamination, or incomplete coverage. BranchManager enforces strict QC filtering before novelty scoring.

• Facts vs. Recommendations are Kept Distinct: Factual states (e.g., physical DNA availability, confirmed genome presence, QC status) are maintained independently from recommendation states (e.g., Primary vs. Backup nomination). Recommendation flags never modify factual records automatically.

• Fail-Closed Safety Principles: Selection-critical processes fail closed. An incomplete tree build, unverified quality score, or broken novelty calculation will halt report publication rather than generating superficially valid recommendations.

• Novelty is Relative to YOUR Project: Novelty metrics specifically measure how unique an isolate is compared to what your project or lab has already characterised, helping to prevent redundant sequencing of strains already in your freezer.

3. The Workflow Stages ("Office Vocabulary")

BranchManager uses an "office workflow" vocabulary to guide users through the multi-step candidate selection lifecycle from initial delivery to annual reporting:

Workflow Stage	Scientific Purpose	Key Outputs & Deliverables
Mailroom	Inventories physical AB1 chromatogram deliveries, reconciles read IDs, and verifies primer metadata.	Batch map (ab1_map.tsv), read inventory, missing/duplicate read reports.
Interview	Standalone chromatogram QC and resequencing triage without project database modifications.	QC assembly report, resequencing recommendations, failed trace manifests.
Onboarding	Validates metadata ledgers, confirms isolate ownership, and enforces unique project ID mapping.	Normalised input datasets, partner ledger validation logs.
Paper Trail & Merge	Converts raw AB1/FASTQ traces, trims low-quality ends, masks low-confidence bases, and merges multi-primer reads.	Assembled FASTA, per-base error profiles, trace visual reports, manual review templates.
Filing Cabinet	Registers cultured baseline datasets (e.g., Hungate) and constructs initial reference backbone trees.	Baseline taxonomy tables, initial tree structures, baseline identity maps.
Performance Review	Classifies marker sequences, screens chimeras, scores multi-lens novelty, and updates phylogenetic trees.	Sequence audit table, novelty summary, iTOL metadata strips, local clade figures.
Hiring Panel	Proposes group-level primary and backup isolate candidates based on current pangenome targets.	Selection summaries, primary/backup diversity plans, dashboard interface.
Quarterly Review	Reconsiders full collections across partner batches to select budgeted genome sequencing tranches.	Tranche nomination lists, marginal distance audits, updated clade visual maps.
Status & Records	Imports factual lab progress (e.g., extraction success, completed WGS assemblies, ANI results).	Updated genome collections, factual isolate status logs, ANI quality gates.
Annual Report	Generates cumulative, audit-ready marker-to-genome close-out reports for institutional stakeholders.	Executive HTML dashboard, complete historical audit trail, ledger close-outs.

4. Key System Tools & Operational Modules

Unified Golden-Path Runner (assistant)

The primary standard execution route. It orchestrates Onboarding, Paper Trail quality control, Performance Review novelty calculations, and Hiring Panel candidate selection in an integrated sequence with zero manual intermediate handling.

Chromatogram Quality Control Engine (mailroom, interview, paper-trail)

Processes raw Sanger sequencing files. Performs Phred quality score evaluation, end-trimming, internal base masking, dye-channel peak analysis for mixed/heterozygous positions, and orientation-aware multi-read consensus assembly. Sequences failing QC are isolated in a dedicated triage folder with detailed diagnostic reason codes.

Baseline & Reference Classifier (filing-cabinet)

Loads physical baseline datasets (such as cultured strain biobanks) to anchor phylogenetic trees and establish the initial novelty threshold. Supports multi-database taxonomic reporting (GTDB, SILVA, GG2, NCBI).

Novelty Scoring & Selection Engine (performance-review, quarterly-review)

Evaluates incoming candidate datasets against the accumulating project database. Calculates neighbourhood crowding metrics, identifies baseline-redundant strains (≥99.8% identity across ≥95% alignment), and applies farthest-first phylogenetic branch distance calculations to pick optimal isolate subsets.

Focused Taxonomy Slicing & Tree Styling (org-chart, label-maker)

Allows researchers to instantly extract specific taxons (e.g., domain Archaea or family Lachnospiraceae) from existing alignments to generate high-resolution focused phylogenetic trees and customise interactive iTOL visual metadata strips without recomputing alignments.

5. Candidate Selection & Decision Categories

BranchManager replaces subjective manual picking with a transparent, evidence-based selection model. Decisions are presented clearly for expert-human review:

Decision Category	Status Type	Scientific Rationale & Use Case
PRIORITISE - SET PRIMARY	Primary Nomination	High-value candidate filling an unrepresented species gap or missing pangenome slot. Highest priority for WGS.
RESERVE - SET BACKUP	Backup Candidate	Phylogenetically diverse strain retained as a secondary option in case the primary candidate fails DNA extraction or library preparation.
EXCLUDE - BASELINE REDUNDANT	Excluded Strain	Strain sharing ≥99.8% 16S identity over ≥95% coverage with an already-cultured baseline strain. Omitted to prevent redundant sequencing.
REVIEW - PANGENOME BOUNDARY	Manual Review	Sequence falls near species identity boundaries; requires manual review to confirm if it represents a distinct novel lineage.
LOWER PRIORITY - TARGET MET	Lower Priority	The maximum target number of genomes for this exact GTDB species has already been satisfied by existing baseline or pending genomes.

6. Core Benefits & Strategic Value

• Sequencing Budget Optimisation: Prevents spending WGS resources on isolates that duplicate strains already characterised in internal or public biobanks.

• Multi-Partner Harmonisation: Provides a single source of truth database across multiple contributing laboratories, ensuring consistent ID mapping and metadata tracking.

• Maximum Diversity Capture: Utilises branch-length algorithms to pick candidates that maximise phylogenetic spread across target clades.

• Full Auditability & Reproducibility: Every candidate recommendation is linked to raw chromatograms, specific quality metrics, component novelty scores, and paginated visual trace reports.

• Scalable Tranche Management: Supports phased sequencing rounds (e.g., quarterly budgets), automatically updating recommendations as newly sequenced genomes become available.

Quick start

For a complete partner submission, use the guarded golden path:

branchmanager assistant \
  --sample-map ab1_mapping.tsv \
  --partner-metadata partner_metadata.tsv \
  --db project.sqlite --dataset QUB_01 \
  --ref gtdb_ssu_reps.fna.gz --ref-name GTDB \
  --chimera-ref silva_gold_ssu.fasta.gz \
  --baseline-fasta hungate.fasta --baseline-dataset Hungate \
  --mwl MWL.csv --threads 10 \
  -o runs/01_QUB_01

assistant runs IT Desk checks, Onboarding, Paper Trail/Merge Meeting, Performance Review, and Hiring Panel in numbered directories. Each stage remains independently runnable:

# 1. Establish the cultured baseline Filing Cabinet and build a reference tree
branchmanager filing-cabinet \
  --fasta hungate.fasta \
  --db project.db \
  --dataset Hungate \
  --ref gtdb_r232_bac_arch_ssu_reps.fna \
  --classify --build-tree \
  -o filing_cabinet_out \
  --threads 10

# 2. Process new sequences — scores novelty against the baseline
branchmanager performance-review \
  --input new_sequences.fasta \
  --db project.db \
  --dataset Batch1 \
  --ref gtdb_r232_bac_arch_ssu_reps.fna \
  --previous-review filing_cabinet_out \
  -o batch1_out \
  --threads 10

# 3. Zoom into a specific taxon
branchmanager org-chart \
  --db project.db \
  --taxon archaea \
  --from-dir filing_cabinet_out \
  -o archaea_out

# 4. Regenerate iTOL files with new grouping options
branchmanager label-maker \
  --db project.db \
  --out filing_cabinet_out \
  --group-phyla archaea \
  --group-phyla "Bacillota,Bacillota_I,Bacillota_A"

---

Core concepts

Novelty is relative to YOUR submitted sequences

All novelty and neighbourhood-density calculations are made against the sequences submitted to BranchManager (the Filing Cabinet baseline plus all prior Performance Review datasets stored in the DB). They are not made against the full external reference (GTDB/SILVA).

This is intentional: BranchManager helps you judge whether a new sequence is worth investigating relative to what your lab or project has already characterised — not relative to all known biology.

Each successive Performance Review extends the reference pool, so novelty scores become increasingly precise as your project grows.

Three-layer tree architecture

The phylogenetic tree is built from three independent layers:

Layer	Source	Shown in iTOL?
1 — Anchors	26 NCBI RefSeq anchor sequences (bundled)	No — invisible topology scaffolding
2 — Filing Cabinet	Your cultured baseline dataset (e.g. Hungate)	Yes
3 — Performance Review	Each new partner marker batch	Yes

Anchor sequences constrain phylum-level topology during MAFFT + FastTree inference, then are pruned from the stored newick. They are never shown in outputs, iTOL files, or novelty scoring.

---

Subcommands

branchmanager filing-cabinet

Load a baseline FASTA dataset, classify against a reference, and build the backbone tree. This is always the first step.

branchmanager filing-cabinet \
  --fasta baseline.fasta \
  --db project.db \
  --dataset Hungate \
  --ref gtdb_r232_bac_arch_ssu_reps.fna \
  --classify --build-tree \
  -o filing_cabinet_out \
  --threads 10 \
  --collapse

Parameter	Required	Default	Description
--fasta	✓	—	Input FASTA file containing the baseline sequences
--db	✓	—	Path to the BranchManager SQLite database (created if absent)
-o / --out		.	Output directory for tree, iTOL files, and reports
--dataset	✓	—	Label stored in the DB (e.g. Hungate)
--ref		—	Reference FASTA (GTDB/SILVA reps) for classification and tree orientation
--classify		off	Classify sequences against --ref and store taxonomy
--build-tree		off	Build the MAFFT + FastTree backbone tree
--taxa		—	TSV mapping reference IDs to lineages (optional when --ref headers contain GTDB lineages)
--taxa-assignments		—	Pre-computed taxonomy for the INPUT sequences (TSV or embedded-lineage FASTA)
--shorten-ids / --no-shorten-ids		--no-shorten-ids	Preserve input headers exactly by default; use --shorten-ids only when compact generated IDs are desired
--collapse		off	Collapse near-identical same-taxonomy sequences into one representative for the tree
--collapse-threshold		99.8	Identity threshold (%) for collapsing
--sequence-domain		bacteria	Domain/profile to process: bacteria, archaea, fungi, or mixed/all/none to disable filtering. Use matching references, previous reviews, and anchors for non-bacterial runs
--anchors		bundled	Custom anchor FASTA for tree topology scaffolding
--threads		4	CPU threads for MAFFT and VSEARCH
--colours		—	CSV mapping sequence IDs to custom hex colours for iTOL (columns: id, colour)
--group-phyla SPEC		—	Group phyla into a single colour in iTOL (repeatable; see Phylum grouping)
--functional		—	TSV file mapping sequence IDs to functional attributes (first column = ID; subsequent columns = attributes). Generates one iTOL file per column (DATASET_BINARY / DATASET_SIMPLEBAR / DATASET_COLORSTRIP).

---

branchmanager mailroom

Prepare the immutable per-batch AB1 map before Onboarding. Mailroom scans every physical chromatogram, extracts ABIF sample/run/instrument and base-call evidence, reconciles supplier sequencing IDs, and reports missing, duplicated, ambiguous, or unmapped reads.

branchmanager mailroom \
  --read-dir UoG/All_AB1 \
  --metadata UoG/supplier_metadata.csv \
  --dataset UoG_01 \
  --forward-primer 63F \
  --reverse-primer 1492R \
  -o UoG/UoG_01

Only use --forward-primer and --reverse-primer after the oligonucleotides have been confirmed from the partner protocol or sequencing submission. Supplier metadata may instead contain a primer column. Mailroom uses embedded ABIF primer names when present, but never treats sequence-motif similarity as proof of the exact primer.

Preferred supplier-metadata columns:

sequencing_id	isolate_number	read	primer	processing_mode
KKX994	SW_0016	Forward	63F	assemble
KKY011	SW_0016	Reverse	1492R	assemble

primer and processing_mode are optional. With --processing-mode auto, paired/multi-read isolates use assemble and single-read isolates use best_read.

Output	Purpose
ab1_map.tsv	Validated one-row-per-read input for Onboarding
ab1_inventory.tsv	Physical file and extracted ABIF metadata inventory
mailroom_report.tsv	Missing, duplicate, ambiguous, unreadable, unmapped, and unresolved-primer findings
mailroom_summary.json	Machine-readable counts and PASS, REVIEW_REQUIRED, or FAIL status
run_manifest.json	Checksums, inputs, outputs, and stage status

Mailroom returns REVIEW_REQUIRED when files reconcile but primer names remain unresolved, and FAIL for structural/file errors. Correct the supplier metadata or confirm primers before Onboarding.

---

branchmanager interview

Run AB1 conversion and QC without Onboarding, taxonomy, tree building, candidate selection, or any project-database access. Interview consumes the complete Mailroom output directory and uses the same versioned QC policy as assistant.

branchmanager interview \
  --mailroom UoG/UoG_01 \
  --screen-ref gtdb_r232_bac_arch_ssu_reps.fna.gz \
  --threads 10 \
  -o UoG/UoG_01_interview

--screen-ref is optional and only checks whether separate primer reads from the same isolate have discordant taxonomic hits. A disagreement produces PASS_WITH_WARNINGS / MANUAL_REVIEW unless another QC criterion fails. The common comparison rank and each read_id=taxon assignment are written to the concordance, assembly, recommendation, and review tables. Without it, all chromatogram, trimming, assembly, quality, ambiguity, mixed-peak, and resequencing outputs are still produced.

Interview accepts either the Mailroom directory or its ab1_map.tsv. It requires the accompanying mailroom_summary.json, rejects a Mailroom FAIL, and stops on REVIEW_REQUIRED unless the batch has been reviewed and --allow-mailroom-review is supplied explicitly.

The principal hand-off files are assembled.fasta, assembly_report.tsv, marker_review_template.tsv, and resequence_recommendations.tsv. Once any manual-review decisions are completed, these can enter the full workflow without repeating chromatogram processing:

branchmanager assistant \
  --fasta UoG/UoG_01_interview/assembled.fasta \
  --marker-qc UoG/UoG_01_interview/assembly_report.tsv \
  --marker-review UoG/UoG_01_interview/marker_review_template.tsv \
  --partner-metadata Sequence_Metadata.csv \
  --db project.sqlite --dataset UoG_01 --ref gtdb_ssu_reps.fna.gz \
  -o runs/UoG_01

Do not supply the blank review template to Assistant. Add decisions only for the listed PASS_WITH_WARNINGS markers first; omit --marker-review when the template contains no rows.

---

branchmanager onboarding

Every partner dataset is onboarded from exactly one marker source. AB1 submissions use a sample map; FASTA-only submissions use the FASTA directly. Both paths validate sequence IDs against the same cumulative project metadata ledger.

# AB1 or multiple primer reads
branchmanager onboarding \
  --sample-map UoG/UoG_01/ab1_map.tsv \
  --read-dir UoG_01_AB1/ \
  --partner-metadata project_partner_metadata.tsv \
  --partner-id UoG --dataset UoG_01 \
  -o UoG_01/01_onboarding

# Partner-supplied marker FASTA; .gz is accepted
branchmanager onboarding \
  --fasta QUB_01_16S.fasta.gz \
  --partner-metadata project_partner_metadata.tsv \
  --partner-id QUB --dataset QUB_01 \
  -o QUB_01/01_onboarding

FASTA Onboarding writes normalised_input.fasta. It does not invent chromatogram quality evidence. Supply a reviewed --marker-qc table to Performance Review where one exists, or use --accept-unverified-marker-qc as an explicit audited acceptance; unverified marker evidence remains visible in the reports.

Partner acronyms and submission labels are separate. For example, UoG_01 and UoG_02 are distinct --dataset values, while both sets of ledger rows use partner_id=UoG. Sequence IDs must be unique and stable across the entire project.

Keep exactly one cumulative metadata ledger with one row per durable isolate/marker ID. It owns partner_id, selected_for_genome_sequencing, and already_sequenced; chromatogram run IDs do not belong in it. Keep a separate read map under each partner/batch, for example UoG/UoG_01/ab1_mapping.csv, which maps every physical trace filename to its durable isolate ID, direction, and processing_mode. Duplicate ledger IDs, partner mismatches, missing traces, and traces assigned to more than one isolate are Onboarding errors.

Onboarding reports only actionable errors and warnings. Normalised submission/read-map tables omit optional columns that are empty throughout the submitted batch.

---

branchmanager paper-trail (AB1 conversion and Merge Meeting)

Convert Sanger AB1 chromatograms or already-basecalled primer reads into a trimmed FASTA for the Performance Review. When multiple reads belong to the same isolate, for example 27F and 907R, BranchManager orients reverse-primer reads and builds a quality-aware overlap consensus.

branchmanager paper-trail \
  --input ab1_reads/ \
  -o paper_trail_out \
  --min-quality 20 \
  --min-mean-quality 20 \
  --min-length 800 \
  --min-overlap 40

branchmanager performance-review \
  --input paper_trail_out/assembled.fasta \
  --partner-metadata new_sequences_metadata.tsv \
  --db project.db \
  --dataset Batch1 \
  --ref gtdb_r232_bac_arch_ssu_reps.fna \
  -o batch1_out

If filenames are structured like Iso001_27F.ab1 and Iso001_907R.ab1, BranchManager can infer the read information. Production partner submissions should nevertheless use one explicit --sample-map per batch, preferably with one row per physical read:

sequence_id	dataset	read_file	primer	direction	processing_mode
Iso001	UoG_01	well_A01.ab1	27F	forward	assemble
Iso001	UoG_01	well_A02.ab1	907R	reverse	assemble
Iso002	UoG_01	well_B01.ab1	27F	forward	best_read

Multiple rows may share the same sequence_id when an isolate has multiple primer reads. Relative paths are resolved from the batch-map location. A wide one-row-per-isolate representation is also accepted when laboratories naturally supply separate primer columns:

isolate_id	27F	907R	processing_mode
Iso001	well_A01.ab1	well_A02.ab1	assemble
Iso002	well_B01.ab1	well_B02.ab1	best_read

Run it with:

branchmanager paper-trail --sample-map sample_reads.tsv -o paper_trail_out

Onboarding writes normalised_read_map.tsv as the validated hand-off to Paper Trail. It is a generated run artefact, not another source ledger and not a file users should maintain.

assemble/merge/consensus orients the primer reads and tries to build one longer sequence. best_read/highest_quality/select_best/independent converts each read independently and writes only the highest-quality passing read for that isolate.

BranchManager applies the same versioned Sanger QC policy in interview, paper-trail, and assistant:

• --min-quality is a Phred cutoff used for Mott-style end trimming.
• --mask-quality masks internal bases below Q20 to N before assembly; masking is reported but does not by itself force manual review.
• Primer sequences are removed only after a confident IUPAC-aware leading match; customise them with repeatable --primer-sequence NAME=SEQUENCE.
• Mixed chromatogram peaks are measured from PLOC and four DATA9-DATA12 dye channels. --secondary-peak-ratio and --max-mixed-peak-percent control the retained-region gate.
• Overlap consensus bases use posterior base probabilities from both Phred observations; qualities are not added as if they were independent certainty scores.
• --screen-ref optionally classifies each primer read independently. Family/genus discordance triggers explicit manual review and reports the differing read-level assignments; it is not a QC failure by itself.
• Final sequences must be at least 800 bp. Individual primer reads may be at least 300 bp so complementary reads can assemble into a valid final marker.
• Ambiguous bases use a tiered rule: more than 3% requires manual review and more than 5% fails QC by default.
• --min-mean-quality, expected-error limits, ambiguity, internal low-quality runs, mixed peaks, and overlap conflict density determine PASS_HIGH_CONFIDENCE, PASS_WITH_WARNINGS, or FAIL_QC.
• Final FAIL_QC sequences are withheld from assembled.fasta and listed as RESEQUENCE.
• Final FAIL_QC sequences and failed reads are retained under failed_qc_sequences/ for manual review.
• PASS_WITH_WARNINGS sequences are included but listed as MANUAL_REVIEW.
• Visual reports are automatically split into numbered PNG pages before they exceed 2400 pixels high. Adjust the ceiling with --max-report-image-height (minimum 600 pixels).

The defaults follow Phred error probabilities and published Sanger guidance that Q20 is a reliable base call and that, as a general rule, trimmed sequence should contain less than 5% ambiguous bases. The 3% review boundary is deliberately more conservative for phylogenetic candidate selection. See Ewing and Green 1998 and Crossley et al. 2020.

Outputs:

File	Description
assembled.fasta	One trimmed/assembled sequence per isolate, suitable for branchmanager performance-review --input
failed_qc_sequences/failed_final_sequences.fasta	Final isolate consensus/best-read sequences that failed final QC
failed_qc_sequences/failed_read_sequences.fasta	Individual read sequences that failed read-level QC, oriented when direction is known
failed_qc_sequences/failed_qc_manifest.tsv	One row per failed isolate, combining the final outcome with read IDs, source files, read-failure evidence, and the best recoverable read
failed_qc_sequences/failed_read_manifest.tsv	One row per failed physical read with trimming and QC evidence; kept separate to avoid duplicate isolate rows
failed_qc_sequences/failed_qc_guide.txt	Plain-language explanation of failure reason codes and the distinction between isolate- and read-level records
trimmed_oriented_reads.fasta	Individual reads after quality trimming and primer-direction orientation
raw_reads.fasta	Raw base calls extracted from AB1 or input sequence files
read_qc.tsv	Per-read trimming, quality, expected error, length, and filter status
per_base_error.tsv	Per-base quality/error probability table with left/right trim and retained-base status
visual_reports/read_error_profiles/read_error_profiles_page_*.png	Paginated per-read quality/error profiles showing retained trim windows
visual_reports/trace_chromatograms/trace_chromatograms_page_*.png	Paginated dye-channel chromatograms with retained windows and mixed-peak evidence
read_taxonomy_concordance.tsv	Independent primer-read classifications, common comparison rank, read_id=taxon assignments, and within-isolate agreement when --screen-ref is used
assembly_report.tsv	Per-isolate assembly status, overlap identity, conflicts, unmerged reads, contributing read IDs, and any read-level taxonomy disagreement with the compared rank and assignments
assembly_read_placements.tsv	Per-read consensus coordinates, aligned blocks, and whether the read contributed to the selected consensus
resequence_recommendations.tsv	Per-isolate ACCEPT, MANUAL_REVIEW, or RESEQUENCE decision with reason codes, taxonomy-concordance evidence, and suggested action
marker_review_template.tsv	Pre-filled rows for manual-review markers, including conflicting read-level taxonomy assignments; complete decision, reviewer, and notes, then provide it as --marker-review
paper_trail_qc_policy.tsv	Versioned thresholds actually used for the run, for reproducibility
visual_reports/assembly_overviews/assembly_overview_page_*.png	Paginated per-isolate overviews with reads placed on consensus coordinates, contribution status, and assembly diagnostics
visual_reports/visual_report_manifest.tsv	Page index with report type, record range, dimensions, and continuation notes
paper_trail_summary.txt	Short run summary
run_manifest.json / .tsv	Input checksums, software/tool versions, thresholds, stages, outputs, and final status

Supported inputs are .ab1, .abi, .fasta, .fa, .fna, .fastq, and .fq, with .gz accepted for AB1/ABI, FASTA, and FASTQ files.

---

branchmanager performance-review

Process new sequences against the baseline; score novelty and update the phylogenetic tree.

branchmanager performance-review \
  --input new_sequences.fasta \
  --db project.db \
  --dataset Batch1 \
  --ref gtdb_r232_bac_arch_ssu_reps.fna \
  --previous-review filing_cabinet_out \
  -o batch1_out \
  --threads 10 \
  --collapse

Parameter	Required	Default	Description
--input	✓	—	FASTA file of new sequences to analyse
--db	✓	—	Path to the BranchManager SQLite database
-o / --out	✓	—	Output directory (assessment TSV, novelty metrics, tree, iTOL files)
--dataset	✓	—	Label for this batch (used in iTOL dataset-membership strip)
--ref		—	Reference FASTA for classification and tree orientation
--previous-review		—	Previous Filing Cabinet or Performance Review output directory used to seed an incremental MSA. When the same -o directory is reused, tree/current_alignment.fasta is detected automatically
--taxa		—	TSV mapping reference IDs to lineages
--taxa-assignments		—	Pre-computed taxonomy for the input sequences
--partner-metadata / --sequencing-metadata	Performance Review	—	Cumulative CSV/TSV ledger with sequence IDs, partner acronyms, optional selected_for_genome_sequencing, and required already_sequenced status
--shorten-ids / --no-shorten-ids		--no-shorten-ids	Preserve input headers exactly by default; use --shorten-ids only when compact generated IDs are desired
--min-len		800	Minimum sequence length to retain (bp)
--max-n-percent		5.0	Maximum percentage of ambiguous (N) bases allowed; matches Paper Trail/Interview final-sequence QC
--marker-qc		auto	Paper Trail/Merge Meeting assembly_report.tsv; auto-discovered beside the FASTA
--marker-review		—	Reviewed accept/reject decisions for warning or unverified marker evidence
--accept-unverified-marker-qc		off	Explicit, audited acceptance of independently validated FASTA without Paper Trail provenance
--chimera-ref		primary ref	Curated chimera-free marker reference used by reference UCHIME
--skip-chimera-check		off	Explicitly omit UCHIME; affected isolates remain review-required
--collapse		off	Collapse near-identical same-taxonomy sequences for the tree
--collapse-threshold		99.8	Identity threshold (%) for collapsing
--sequence-domain		bacteria	Domain/profile to process: bacteria, archaea, fungi, or mixed/all/none to disable filtering. Run archaea/fungi separately with matching --ref, --alt-ref, --baseline-fasta, --previous-review, and --anchors
--phylum		—	Filter iTOL output to a specific phylum (does not affect novelty scoring)
--target		—	FASTA to measure novelty against instead of the DB
--baseline-fasta		—	Baseline/context FASTA to load before evaluating, e.g. Hungate
--baseline-dataset		Baseline	Dataset label for --baseline-fasta and default cultured-baseline novelty pool
--novelty-baseline-dataset		—	Existing DB dataset to include in the baseline novelty pool; repeatable
--baseline-shorten-ids / --no-baseline-shorten-ids		--no-baseline-shorten-ids	Preserve baseline IDs exactly by default
--force-rebuild		off	Force a full tree rebuild from scratch
--anchors		bundled	Custom anchor FASTA for tree scaffolding
--threads		4	CPU threads
--neighbourhood-format		png	Local-clade figure format; PNG is the only supported output
--user-colours		—	CSV mapping sequence IDs to custom hex colours for iTOL
--group-phyla SPEC		—	Group phyla into a single colour in iTOL (repeatable)
--functional		—	TSV file mapping sequence IDs to functional attributes (first column = ID; subsequent columns = attributes). Generates one iTOL file per column (DATASET_BINARY / DATASET_SIMPLEBAR / DATASET_COLORSTRIP).

Rolling project runs

Use one persistent SQLite database for the project and give each partner submission a unique --dataset label. The input FASTA may contain only the genuinely new isolates; --partner-metadata may be a cumulative table containing both new and previously stored partner IDs, so changed genome-selection flags are applied before the collection is reassessed.

For an auditable history, use a new output directory for every Performance Review and pass the preceding review with --previous-review. Reusing the same output directory is also supported and automatically resumes from tree/current_alignment.fasta, but replaces that directory's previous reports. Keep reference-database versions, domain, QC settings, collapse settings, anchors, and tree method constant when comparing reviews. Do not use --force-rebuild unless a deliberate full reanalysis is required.

Sequence IDs are persistent keys. Resubmitting an existing ID with the same sequence is allowed; resubmitting it with a different sequence is rejected rather than silently changing the project record. External-reference and cultured-baseline evidence should remain stable when inputs and settings are unchanged. Project crowding, nearest-project hit, genome-collection coverage, pangenome gap, selection-set role, and recommendation are deliberately recalculated because new isolates and updated WGS commitments change those questions.

Incremental runs retain the old MSA columns and add new sequences with MAFFT. The tree builder still estimates a new tree from the expanded alignment, so small changes in old topology or branch lengths remain scientifically possible. Use the preceding Performance Review directory as --previous-review, keep parameters/reference versions fixed, inspect assessment/decision_changes.tsv, and reserve --force-rebuild for deliberate recalibration. Strictly fixed old placements require a validated fixed-backbone phylogenetic-placement workflow and are not claimed by BranchManager.

Domain/profile handling:

BranchManager defaults to --sequence-domain bacteria for Filing Cabinet and Performance Review. This keeps bacterial 16S runs clean by filtering non-bacterial assignments before DB insertion and tree building where taxonomy is available. Process archaea and fungi as separate reviews with their own DB, outputs, baselines, references, and anchors:

branchmanager performance-review ... --sequence-domain archaea --ref archaeal_16s_refs.fasta --previous-review archaea_filing_cabinet -o archaea_review
branchmanager performance-review ... --sequence-domain fungi --ref fungal_its_or_18s_refs.fasta --anchors fungal_anchors.fasta -o fungi_review

Use --sequence-domain mixed when you intentionally want to keep all domains in one run.

---

branchmanager org-chart

Extract all sequences matching a given taxon from the DB and build a focused phylogenetic tree for that group only.

Fast path (recommended): if --from-dir points to a directory containing current_alignment.fasta, sequences are sliced from the pre-built MSA and only FastTree is run — seconds to minutes for any sized group.

Slow path: if no existing alignment is found, a full MAFFT + FastTree build is performed.

# Domain-level (auto-detected keyword)
branchmanager org-chart --db project.db --taxon archaea      --from-dir filing_cabinet_out -o archaea_out
branchmanager org-chart --db project.db --taxon bacteria     --from-dir filing_cabinet_out -o bacteria_out

# Phylum — plain name (rank auto-detected) or GTDB-prefixed
branchmanager org-chart --db project.db --taxon Bacteroidota       --from-dir filing_cabinet_out -o bact_out
branchmanager org-chart --db project.db --taxon p__Bacillota       --from-dir filing_cabinet_out -o firm_out

# Family
branchmanager org-chart --db project.db --taxon f__Lachnospiraceae --from-dir filing_cabinet_out -o lachno_out

# Genus
branchmanager org-chart --db project.db --taxon g__Ruminococcus    --from-dir filing_cabinet_out -o rumino_out

Accepted taxon formats:

Input	Detected rank	Notes
archaea / bacteria	domain	Auto-matched case-insensitively
Bacillota	phylum (fallback)	No prefix → defaults to phylum
p__Bacillota	phylum	GTDB prefix auto-detected
f__Lachnospiraceae	family
g__Ruminococcus	genus
d__Archaea	domain	Explicit GTDB domain prefix

Parameter	Required	Default	Description
--db	✓	—	Path to the BranchManager SQLite database
-o / --out	✓	—	Output directory
--taxon	✓	—	Taxon to extract (see table above)
--rank		auto	Override rank detection: domain / phylum / family / genus / species
--from-dir		—	Existing Filing Cabinet or Performance Review output directory with current_alignment.fasta or tree/current_alignment.fasta (fast path)
--ref		—	Reference FASTA for orientation correction (slow path only)
--anchors		bundled	Custom anchor FASTA
--threads		4	CPU threads
--min-seqs		3	Minimum matching sequences required to build a tree
--no-tree		off	Skip tree building; only write taxonomy TSV and iTOL colour files
--group-phyla SPEC		—	Group phyla into a single colour in iTOL (repeatable)
--functional		—	TSV file mapping sequence IDs to functional attributes (first column = ID; subsequent columns = attributes). Generates one iTOL file per column (DATASET_BINARY / DATASET_SIMPLEBAR / DATASET_COLORSTRIP).

Org Chart outputs:

File	Contents
org_chart_tree.nwk	Focused newick tree (anchor-free, nodes labelled NODE####)
org_chart_alignment.fasta	Filtered alignment slice used for the tree
org_chart_combined_taxonomy.tsv	ID → taxonomy → confidence for matched sequences
org_chart_sequence_list.tsv	ID, taxonomy, confidence, dataset for matched sequences
itol_phylum_colours.itol	iTOL colour strip by phylum
itol_dataset_membership.itol	iTOL strip showing which dataset each sequence came from

---

branchmanager label-maker

Regenerate all iTOL colour files from taxonomy already stored in the DB — without re-classifying or rebuilding the tree. Useful when changing phylum groupings.

branchmanager label-maker \
  --db project.db \
  --out filing_cabinet_out \
  --group-phyla archaea \
  --group-phyla "Bacillota,Bacillota_I,Bacillota_A"

Parameter	Required	Default	Description
--db	✓	—	Path to the BranchManager SQLite database
-o / --out	✓	—	Output directory (use your Filing Cabinet or Performance Review output directory)
--include-datasets		all	Comma-separated dataset names to include
--sequence-domain		—	Include only sequences from this domain/profile
--group-phyla SPEC		—	Group phyla into a single colour (repeatable)
--functional		—	TSV file mapping sequence IDs to functional attributes (first column = ID; subsequent columns = attributes). Generates one iTOL file per column (DATASET_BINARY / DATASET_SIMPLEBAR / DATASET_COLORSTRIP).

---

Phylum grouping for iTOL (--group-phyla)

Available on all four subcommands. Groups multiple phyla into a single colour entry in the iTOL legend.

Spec format	Effect
archaea	All archaeal phyla (detected by d__Archaea domain) → one colour labelled Archaea
bacteria	All bacterial phyla → one colour labelled Bacteria
"Bacillota,Bacillota_I,Bacillota_A"	These three phyla → one colour; label = first name (Bacillota)
"Firmicutes:Bacillota,Bacillota_I"	These two phyla → one colour labelled Firmicutes

Multiple --group-phyla arguments are independent and additive:

branchmanager performance-review ... \
  --group-phyla archaea \
  --group-phyla "Bacillota,Bacillota_I,Bacillota_A" \
  --group-phyla "Bacteroidota,Bacteroidota_A"

---

Multiple dataset stacking

Datasets accumulate in the DB. Each Performance Review sees the growing baseline and project collection:

# Step 1 — baseline
branchmanager filing-cabinet --fasta hungate.fasta --db project.db --dataset Hungate ...

# Step 2 — first batch: novelty scored against Hungate
branchmanager performance-review --input batch1.fasta --db project.db --dataset Batch1 ...

# Step 3 — second batch: novelty scored against Hungate + Batch1
branchmanager performance-review --input batch2.fasta --db project.db --dataset Batch2 ...

---

Clustering and tree redundancy reduction (--collapse)

When --collapse is enabled, BranchManager groups sequences sharing ≥ --collapse-threshold identity and the same taxonomy, keeping only one cluster representative per group for tree building.

Column Explanation

Column	Meaning
ID	User-supplied sequence ID (preserved unless ID shortening is explicitly requested)
Taxonomy	Full GTDB lineage assigned by the classifier
BestHit	Closest reference genome accession (VSEARCH best hit)
ClassificationIdentity	% identity to the BestHit reference (VSEARCH alignment)
ClassificationConfidence	Confidence of the taxonomy assignment (from the taxa TSV); NA when taxonomy is parsed directly from reference FASTA headers (no confidence column available)
NearestHit	Closest sequence among YOUR previously submitted sequences (Filing Cabinet + prior Performance Reviews)
NearestIdentity	% identity to NearestHit
MatchesGE99 / GE97 / GE95	Count of YOUR submitted sequences within 99% / 97% / 95% identity
NoveltyScore	100 − NearestIdentity (higher = more novel vs. your collection)
Crowding	Neighbourhood density: crowded / moderate / sparse based on how many of your sequences are nearby
SequencingPriority	HIGH / MEDIUM / LOW — suggested priority for follow-up based on novelty + crowding
Reference*	Parallel nearest hit, identity, score, crowding, and priority against the selected external reference FASTA, usually GTDB
InTree	Yes = entered the phylogenetic tree; No = excluded (see ClusterRepresentative)
ClusterRepresentative	self = this sequence is in the tree; an ID = represented by that leaf after optional collapse; duplicate = retained isolate omitted from the final tree as an exact marker duplicate
ClusterSize	Total sequences in this cluster (1 = singleton)
ClusteredMembers	Semicolon-separated IDs of OTHER sequences collapsed under this representative
PlacementFlags	Warnings: LOW_CLASSIFICATION_IDENTITY, LOW_NEAREST_IDENTITY, NOVEL_BUT_ASSIGNED, etc.

novelty_metrics.tsv

This table keeps the novelty lenses separate:

• Hungate*: nearest hit, identity, query coverage, novelty, and density against priority Hungate baselines.
• SecondaryBaseline*: the same evidence against secondary cultured-rumen baseline datasets.
• CulturedRumen*: closest evidence across Hungate and secondary baseline tiers combined.
• Project*: nearest hit and density against all partner candidates accumulated across runs, including the current collection but excluding each query's self-hit.
• Reference*: nearest hit and density against the selected external reference FASTA supplied with --ref, usually GTDB.
• GenomeCollection* / Pangenome*: rolling genome coverage. Every baseline isolate counts because its genome is available; partner isolates count once selected. The default target is nine committed genomes per exact GTDB species or unresolved local clade.
• SelectionGroupType: HUNGATE_BASELINE_EXTENSION means a Hungate genome anchors the exact GTDB species, SECONDARY_BASELINE_EXTENSION means only secondary cultured-rumen genomes anchor it, and CANDIDATE_ONLY_GROUP has no baseline-genome anchor.
• BaselineExtension*: membership gate for a baseline-pangenome extension. The candidate and nearest cultured-rumen baseline must have the same exact GTDB species, with >=98.65% 16S identity across >=95% of the query by default. A failed gate is PANGENOME_BOUNDARY_REVIEW, not an automatic rejection: it may represent a separate candidate lineage.
• BaselineRedundancy*: hard eligibility gate for uncommitted candidates. By default, a nearest cultured-rumen hit at >=99.8% identity across >=95% of the query is reported as BASELINE_REDUNDANT and receives no panel rank.
• SelectionDiversityDistance: marginal patristic distance from baseline and already committed genome markers when tree distances are available. This drives farthest-first diversity capture; cultured-baseline, project, and GTDB-reference divergence provide the fallback, with MWL evidence used as an additional tie-break signal.

A species represented by one baseline genome therefore starts at 1/9, leaving eight primary genome slots. Eligible partner isolates are ranked to extend that baseline pangenome; additional ranked isolates can be retained as extraction-failure/diversity backups. MWL evidence can order eligible isolates but cannot override the species/coverage boundary or the near-identical baseline exclusion.

---

All output files

Filing Cabinet outputs

File	Description
current_tree.nwk	Backbone phylogenetic tree (newick, nodes labelled NODE####)
current_alignment.fasta	Full multiple sequence alignment (includes anchor sequences)
filing_cabinet_combined_taxonomy.tsv	ID → taxonomy → confidence for all registered baseline sequences
taxonomy.tsv	Raw classifier output (ID, best-hit, identity, taxonomy, confidence)
itol_phylum_colours.itol	iTOL colour strip by phylum
itol_family_colours.itol	iTOL colour strip by family
itol_genus_colours.itol	iTOL colour strip by genus
filing_cabinet_dataset.itol	iTOL strip marking all sequences with the dataset colour
filing_cabinet_id_map.tsv	Short ID → original FASTA header mapping
filing_cabinet_collapsed_map.tsv	Cluster representative → taxonomy → count
filing_cabinet_collapsed_members.tsv	Member → representative mapping
tree_build_warnings.tsv	Warnings about sequence quality or alignment issues
tree_orientation_summary.tsv	Per-sequence orientation audit (forward / RC / unknown)
OUTPUT_EXPLANATIONS.tsv	Manifest describing each output file

Performance Review and Hiring Panel outputs

File	Description
performance_review_dashboard.html	Compact linked view of the current Hiring Panel recommendations
assessment/sequence_assessment.tsv	Current-dataset full audit table. Per-sequence novelty, taxonomy, crowding, priority, clustering, MWL matches, and placement flags for only the dataset supplied to this Assistant/Performance Review run
assessment/selection_summary.tsv	Current-dataset SAB decision table. Recommendation, evidence quality, component novelty evidence, same-species genome coverage, candidate-set role, rationale, and local-tree link
assessment/sequencing_sets.tsv	Current-dataset nine-member diversity plan. Prior datasets and baselines inform coverage and diversity, but only current input candidates receive rows in this run
assessment/novelty_metrics.tsv	Current-dataset novelty and crowding summary; searches use the cumulative baseline and project collection as comparison context
assessment/neighbourhoods/clade_*.png	Labelled local phylogenetic neighbourhoods (default). Nearby assessed isolates are grouped into one figure rather than duplicated across figures
assessment/neighbourhoods/clade_*_pairwise_pident.tsv	Complete long-form MSA percent-identity table for every pair of displayed tree leaves, including compared-column counts and the identity definition
assessment/neighbourhoods/neighbourhood_manifest.tsv	Maps every assessed sequence to its image and pident table, including the P1 identity anchor, forced nearest-baseline hits, assessed isolates, baseline leaves, and already-sequenced genomes shown
assessment/mwl_matches.tsv	Most Wanted List hits when --mwl is supplied
assessment/cluster_summary.tsv	Cluster-level prioritisation report when clustering/tree reports are available
assessment/clusters.csv	One consolidated membership table containing all clusters and all member isolates; replaces the former one-file-per-cluster directory
assessment/backup_candidates.tsv	Ranked alternative isolates within clusters when a primary candidate cannot be sequenced
baseline/baseline_hits.tsv	Nearest-hit report against Hungate or other provided baseline datasets
baseline/nearest_baseline_hits_raw.tsv	Raw nearest-hit output used for baseline novelty scoring
taxonomy/<DB>.tsv	Taxonomic assignment report for each reference database, e.g. taxonomy/GTDB.tsv, taxonomy/GG2.tsv
taxonomy/all_databases.tsv	Multi-database assignment summary
taxonomy/tree_taxonomy.tsv	ID → taxonomy → confidence table used for tree/iTOL metadata
tree/current_tree.nwk	Cumulative project-context tree incorporating baseline, prior partner datasets, and the current dataset
tree/current_alignment.fasta	Cumulative MSA used to build the project-context tree
tree/tree_build_warnings.tsv	Tree-quality warnings
itol/phylum.itol	iTOL colour strip by phylum
itol/family.itol	iTOL colour strip by family
itol/genus.itol	iTOL colour strip by genus
itol/dataset_membership.itol	iTOL strip showing which dataset each sequence belongs to
itol/novelty.itol	iTOL strip showing nearest-hit novelty
ids/user_id_map.tsv	Short ID → original FASTA header mapping for this run
intermediate/	QC, optional collapse, and classifier scratch outputs retained for debugging; exact marker duplicates retain separate isolate IDs in assessment reports
logs/branchmanager.log	Run log

BranchManager keeps iTOL DATASET_COLORSTRIP files, one per metadata type. Older TREE_COLORS branch/range files and symbol-strip variants are removed because they encoded the same metadata in additional visual styles.

selection_summary.tsv

This is the board-facing table for scientific advisory board discussions. It avoids an opaque adjusted score and keeps the evidence components visible: cultured-baseline novelty, rolling project coverage, GTDB-reference divergence, rank-aware MWL evidence, exact same-species genome counts, marker evidence quality, candidate-set role, and local-tree figure.

SelectedForGenomeSequencing and GenomeAlreadySequenced keep pending commitments distinct from completed genomes. SelectedPendingGenomesSameSpecies contributes to CommittedGenomesSameSpecies but not to nearest-available-genome identity.

The decision labels are:

Decision	Meaning
PRIORITISE - SET PRIMARY	Fills one of the missing genomes needed to reach the per-species pangenome target
RESERVE - SET BACKUP	Additional phylogenetically spread isolate retained for extraction failure and strain-diversity capture
STRONG CANDIDATE / SECONDARY CANDIDATE	Useful evidence, but not currently assigned a primary/backup place in the working set
SECONDARY - STRAIN DIVERSITY	The numerical target is met, but the isolate adds a ranked diversity direction not represented by baseline/committed markers
REVIEW - PANGENOME BOUNDARY	The species, identity, or coverage evidence does not justify placing the isolate in the baseline-pangenome extension; review it as a possible separate lineage
EXCLUDE - BASELINE REDUNDANT	>=99.8% identity to a cultured baseline marker across >=95% query coverage; retained in reports but excluded from ranking
LOWER PRIORITY - TARGET MET	Available genomes plus committed pending selections meet the requested exact-GTDB-species target
REVIEW BEFORE SELECTION	Marker classification, QC, disagreement, or required comparison evidence needs review
ALREADY SELECTED - GENOME PENDING	The isolate is already committed to sequencing but no usable genome is available yet
ALREADY SEQUENCED	The current isolate already has an available genome and is not a new sequencing recommendation

Project-wide expansion rounds

The default Performance Review targets nine committed genomes per exact GTDB species, or per local phylogenetic clade when species assignment is unresolved. It writes up to nine ranked eligible partner candidates per group. Baseline and completed/pending partner genomes determine the remaining target gap, while candidates outside that gap remain ranked as backups or post-target diversity options.

Near-identical cultured-baseline matches are excluded only when both identity and alignment extent support the claim: >=99.8% marker identity and >=95% query coverage by default. This prevents a short high-identity fragment from being treated as redundant. Diversity is ranked first using marginal tree distance from baseline and committed genomes where available, then baseline/project/reference divergence and phylogenetic isolation. MWL rank and score add priority among otherwise comparable diversity choices but never override the baseline-redundancy gate.

After partner submissions have accumulated and sequencing statuses have been updated, run a Quarterly Review to choose the next tranche across the whole project:

branchmanager quarterly-review \
  --db project.sqlite \
  --genome-budget 24 \
  --backups-per-primary 1 \
  --tree runs/final/tree/current_tree.nwk \
  --alignment runs/final/tree/current_alignment.fasta \
  --partner-metadata partner_status.tsv \
  -o quarterly_review_01

--genome-budget is the number of new primary nominations, not including backups. Quarterly Review fills remaining nine-genome coverage gaps first, filters baseline-redundant candidates, then balances diversity across species. Within those tiers it uses marginal tree distance where available, nearest-genome marker identity, cultured-baseline novelty, GTDB-reference context, phylogenetic isolation, and MWL evidence. When --alignment is supplied, nearest available-genome identity is recomputed against the current genome collection, so newly completed genomes affect the next round immediately. It does not change already_sequenced; recommendations become available genomes only after metadata is updated in a later round.

Every current Performance Review stores its normalised assessment rows in the project database. For an older review made before assessment snapshots were introduced, import its full audit table once:

branchmanager quarterly-review \
  --db project.sqlite \
  --assessment runs/01_Batch1/assessment/sequence_assessment.tsv \
  --genome-budget 24 \
  --tree runs/01_Batch1/tree/current_tree.nwk \
  -o quarterly_review_01

Quarterly Review outputs are:

File	Description
next_genome_set.tsv	Current PRIMARY nominations and their BACKUP isolates
quarterly_review_summary.tsv	Full audit containing all candidates, already-sequenced isolates, review exclusions, tiers, marginal distances, and reasons
quarterly_review_manifest.tsv	Round parameters, recommendation counts, and scientific scope
neighbourhoods/clade_*.png	Round-specific local-clade figures whose stars show this Quarterly Review's primary and backup nominations

Quarterly Review is marker-gene decision support. Once genomes are available, strain-level expansion should be checked with ANI, phylogenomics, and pangenome structure rather than treating 16S distance as a strain boundary.

sequence_assessment.tsv columns

Column	Meaning
ID	Sequence ID supplied by the user; IDs are preserved unless shortening is explicitly requested
GTDBTaxonomy	Full authoritative GTDB lineage assigned by the classifier
GTDBClassificationHit	Closest GTDB reference accession used for classification
GTDBClassificationIdentity	% identity to the classification hit
GTDBClassificationConfidence	Confidence of the GTDB assignment
BaselineNearestHit	Closest sequence in the cultured baseline, e.g. Hungate
BaselineNearestHitDataset	Dataset label for that baseline hit
BaselineNearestIdentity	% identity to the cultured-baseline hit
BaselineMatchesGE99 / GE97 / GE95	Count of baseline sequences within 99% / 97% / 95% identity
BaselineNoveltyScore	0–100 score; higher = more divergent and less crowded versus cultured isolates
BaselineCrowding	Baseline neighbourhood density: crowded / moderate / sparse / isolated
BaselinePriority	HIGH / MEDIUM / LOW heuristic from baseline novelty and density
Project*	Parallel novelty hit, identity, score, crowding, and priority against the rolling partner-candidate collection, excluding self
GTDBReference*	Parallel nearest hit, identity, score, and crowding against the authoritative GTDB reference
PartnerID	Partner acronym loaded from --partner-metadata, e.g. QUB or UoG
SelectedForGenomeSequencing	Whether the project has committed this isolate to sequencing; the genome may still be pending
GenomeAlreadySequenced	Whether the current partner isolate already has an available genome; a recommendation alone does not set this field
GenomeCollection*	Nearest hit and density across all baseline genomes plus already-sequenced partner genomes
BaselineGenomesSameAssessmentSpecies / SequencedPartnerGenomesSameAssessmentSpecies	Existing exact-species genome coverage by source; assessment species means GTDB for bacterial/archaeal Performance Reviews
CommittedGenomesSameAssessmentSpecies / SelectedPendingGenomesSameAssessmentSpecies / PangenomeTarget / PangenomeGap	Rolling completed-genome coverage, pending commitments, and the number of additional selections required
SequencingSet*	Stable group ID and proposed primary/backup/alternate action for the isolate
LocalNeighbourhoodFigure	Relative path to the local-clade PNG containing this assessed sequence
TreeContextLeafCount	Number of tree leaves displayed in that figure; this is display context, not a crowding score
AssessedSequencesInTreeContext	Number of current assessed isolates sharing the figure
InTree	Yes = in tree; No = excluded (see ClusterRepresentative)
ClusterRepresentative	self, a representative ID, or duplicate
ClusterSize	Sequences in this cluster
ClusteredMembers	Other sequences collapsed under this representative
PlacementFlags	Warning codes (e.g. LOW_CLASSIFICATION_IDENTITY)

---

Reference inputs — which flag to use

Scenario	Recommended flags
Classify against GTDB reps FASTA	--ref gtdb.fna --classify
Input FASTA has embedded GTDB lineages in headers	--taxa-assignments input.fasta
Separate taxonomy TSV for the input sequences	--taxa-assignments assignments.tsv
GTDB FASTA + separate taxonomy TSV for the reference	--ref gtdb.fna --taxa gtdb_taxonomy.tsv --classify

--ref = the reference database used for classification and tree orientation.
--taxa-assignments = pre-computed taxonomy for the input sequences themselves.

Partner sequencing metadata

branchmanager performance-review requires --partner-metadata / --sequencing-metadata. Provide this as a simple sidecar .csv, .tsv, or gzipped CSV/TSV alongside the FASTA. It must contain:

• A sequence ID column such as sequence_id, isolate_id, sample_id, or ID. Values may match IDs in the current FASTA or partner isolates already stored in the same project SQLite database.
• A partner acronym column such as partner_id, partner, or partner_acronym, with values like QUB or UoG.
• An optional selection-commitment column such as selected_for_genome_sequencing. Use yes after the project has actually committed that isolate to sequencing.
• An already-sequenced/genome-available column such as already_sequenced, genome_available, or genome_sequenced. This column is required and must be explicit yes or no.

Example:

sequence_id	partner_id	selected_for_genome_sequencing	already_sequenced
Iso001	QUB	yes	no
Iso002	UoG	yes	yes
Iso003	UoG	no	no

Set selected_for_genome_sequencing=yes only after the advisory/project decision has been made. It reserves that isolate in rolling pangenome commitments but does not put it in the nearest-genome comparison pool. Set already_sequenced=yes only when a usable genome is genuinely available; that state does enter WGS coverage and nearest-genome calculations. PRIMARY and BACKUP are BranchManager recommendations and do not update either factual state automatically.

Keep one cumulative ledger for the project. It may contain rows from earlier submissions even when the current FASTA contains only new sequences. Each Performance Review refreshes matched stored isolates before recalculating commitments, available-genome coverage, and recommendations.

---

ID shortening

By default, BranchManager preserves the FASTA IDs exactly as supplied. This is especially important for Hungate/baseline datasets and partner-provided isolate IDs.

• --no-shorten-ids (default) — keep supplied IDs.
• --shorten-ids — generate compact IDs when explicitly requested.
• --no-baseline-shorten-ids (default for Performance Review baselines) — keep baseline IDs exactly.
• --baseline-shorten-ids — generate compact baseline IDs when explicitly requested.

When IDs are shortened, BranchManager writes an ID map under ids/. When IDs are preserved, the map is still useful as an audit trail but should normally be identity-to-identity.

---

Anchor sequences

Anchor sequences are invisible scaffolding constraining phylum-level topology during tree construction. They are pruned from the stored newick after each build and never appear in outputs, iTOL files, or novelty scoring.

The bundled anchor set (src/branchmanager/data/reference_anchors.fasta) contains 26 NCBI RefSeq 16S sequences covering major gut/rumen phyla. A companion metadata table (src/branchmanager/data/reference_anchors.tsv) explains what each anchor represents, its rumen relevance, and whether it is a core rumen anchor or broader topology scaffold.

Phylum	Type strain	Accession
Bacillota	Ruminococcus albus 7	NR_025930
Bacillota	Clostridium butyricum ATCC 19398	NR_074545
Bacillota	Bacillus subtilis 168	NR_027552
Bacillota	Lactobacillus acidophilus ATCC 4356	NR_075051
Bacillota	Streptococcus equinus NBRC 12553	NR_113594
Bacillota	Butyrivibrio fibrisolvens D1	NR_044858
Bacteroidota	Bacteroides fragilis ATCC 25285	NR_041386
Bacteroidota	Prevotella bryantii B14	NR_044825
Bacteroidota	Porphyromonas gingivalis ATCC 33277	NR_040847
Pseudomonadota	Escherichia coli K-12	NR_102804
Pseudomonadota	Helicobacter pylori 26695	NR_073694
Pseudomonadota	Wolinella succinogenes DSM 1740	NR_043184
Actinomycetota	Bifidobacterium longum NCC2705	NR_040783
Actinomycetota	Streptomyces griseus ATCC 23345	NR_043823
Spirochaetota	Treponema pallidum str. Nichols	NR_027243
Spirochaetota	Borrelia burgdorferi B31	NR_025890
Fibrobacterota	Fibrobacter succinogenes S85	NR_041558
Fusobacteriota	Fusobacterium nucleatum ATCC 25586	NR_026043
Planctomycetota	Planctomyces maris DSM 8797	NR_043399
Verrucomicrobiota	Akkermansia muciniphila ATCC BAA-835	NR_042817
Thermotogota	Thermotoga maritima MSB8	NR_043084
Deinococcota	Deinococcus radiodurans R1	NR_036779
Cyanobacteriota	Synechococcus elongatus PCC 6301	NR_043317
Archaea (outgroup)	Methanobrevibacter ruminantium M1	NR_044812
Archaea (outgroup)	Methanobacterium thermoautotrophicum Marburg	NR_028243
Archaea (outgroup)	Sulfolobus acidocaldarius DSM 639	NR_043325

This is a pragmatic rumen/gut scaffold rather than a final rumen-only reference panel. The TSV flags which anchors are core rumen representatives and which are broader topology anchors that can be replaced as better vetted rumen-specific sequences become available.

To use a custom anchor file:

branchmanager filing-cabinet --anchors /path/to/my_anchors.fasta ...

Custom anchor headers must begin with BRANCHMANAGER_REF_:

>BRANCHMANAGER_REF_MyPhylum accession=NR_XXXXXX source=SILVA138

To refresh the bundled anchors from NCBI:

python scripts/build_anchor_fasta.py --email your@institution.ac.uk

---

Interpreting outputs

Selection evidence

Metric	Meaning
CulturedGap	LARGE below 97%, MODERATE from 97% to <98.65%, and SMALL at >=98.65% against the cultured baseline
ProjectCoverage	Number of other rolling partner candidates at >=97% identity: uncovered, sparse (1-3), moderate (4-10), or dense (>10)
ReferenceContext	HIGH DIVERGENCE below 94.5%, MODERATE DIVERGENCE from 94.5% to <98.65%, or CLOSE REFERENCE at >=98.65%
GenomeCoverage	Exact same-GTDB-species available-genome count and remaining target gap; nearest-genome identity is shown separately as context
EvidenceQuality	Whether classification identity and placement warnings support using the marker result for selection

The 94.5% and 98.65% values are commonly used full-length 16S heuristics for genus- and species-range divergence. BranchManager uses them as decision-support categories, not as declarations of a new genus or species. Sequence quality, alignment coverage, marker copy variation, phylogenetic placement, and ultimately genome-based analysis still matter.

Threshold basis: Yarza et al. (2014), genus and higher-rank 16S boundaries and Kim et al. (2014), 98.65% 16S/ANI species-demarcation correspondence. Their use in rumen-focused taxonomy is discussed by Henderson et al. (2019), including limitations for unusually diverse genera.

The local-clade images expand around each assessed tree leaf to show a compact context (normally 8-30 leaves). Overlapping contexts are merged when their common clade remains compact. The nearest cultured-baseline hit for each grouped context is forced into the image even when reaching it crosses the normal leaf limit; less informative non-target leaves are pruned first. An orange diamond and [SEQUENCED] mark a confirmed available genome. A filled purple star marks the current sequence-next recommendation ([P1], or another primary if more than one genome is missing); outlined purple stars mark recommended backups ([B2] and later ranks). [ALT] isolates are contextual alternatives, not current recommendations.

Each leaf label reports MSA pident relative to the figure's P1 sequence, and clade_*_pairwise_pident.tsv contains all displayed leaf-to-leaf comparisons. MSA pident is defined as identical A/C/G/T bases divided by alignment columns where both sequences contain an unambiguous A/C/G/T base; ComparableACGTColumns reports the overlap supporting the value. Terminal overhangs, gaps, and ambiguous bases are excluded rather than being misreported as substitutions. Internal nodes are inferred ancestors without observed marker sequences, so BranchManager does not assign pident to them; branch geometry and the scale bar remain modelled substitutions per site. These plots support visual inspection, while quantitative crowding remains the vsearch MatchesGE* count against explicitly named pools.

Placement flags

Flag	Meaning
LOW_CLASSIFICATION_IDENTITY	< 95% identity to best reference hit
LOW_CONFIDENCE	Classification confidence below threshold
LOW_NEAREST_IDENTITY	< 95% identity to nearest sequence in your collection
NOVEL_BUT_ASSIGNED	High novelty score despite a confident taxonomy assignment

Tree quality warnings

The tree is a context plot, not absolute proof of novelty. Always pair tree inspection with sequence_assessment.tsv. Warnings in tree_build_warnings.tsv:

Warning	Meaning
PARTIAL_16S_SEQUENCES	Sequences shorter than 1200 bp reduce placement accuracy
HIGH_N_CONTENT	High ambiguous-base content can distort branch lengths
VERY_SHORT_ALIGNED_FRAGMENTS	Some aligned rows are mostly gaps after MAFFT
MISSING_REFERENCE_ANCHORS	Tree was built without topology scaffolding

If a large unknown sector appears in iTOL, check:

• combined_taxonomy.tsv or filing_cabinet_combined_taxonomy.tsv
• taxonomy.tsv (raw classifier output)
• taxonomy_input_warnings.tsv

---

Operational updates and project close-out

Recommendations are not genome facts. Use status-meeting for laboratory progress and records-update only when genome evidence exists.

# Laboratory/SAB lifecycle changes
branchmanager status-meeting --db project.sqlite --input isolate_status.tsv -o updates/status_01

# Completed assemblies, with automatic 90% completeness / 5% contamination defaults
branchmanager records-update --db project.sqlite --input genome_results.tsv -o updates/genomes_01

# Re-open the complete collection after the initial target-nine selection phase
branchmanager quarterly-review --db project.sqlite --genome-budget 24 \
  --from-dir runs/latest/03_performance_review_hiring_panel -o quarterly_reviews/round_02

# Final cumulative report after sequencing and genome QC
branchmanager annual-report --db project.sqlite -o final_annual_report

isolate_status.tsv requires sequence_id and status; supported states include RECEIVED, TRACE_REVIEW, MARKER_QC_PASSED, PROPOSED, SAB_APPROVED, DNA_EXTRACTION_FAILED, SEQUENCED, GENOME_QC_FAILED, and GENOME_QC_PASSED.

genome_results.tsv requires sequence_id and genome_id/accession. Recommended fields are genome_status, completeness, contamination, gtdb_taxonomy, ani_cluster, and genome_path. A SEQUENCED row with both quality metrics is automatically assigned genome-QC pass/fail using the configured thresholds. Explicit QC pass values that conflict with supplied metrics are rejected. Only QC-passed genomes update the available genome collection.

Every Performance Review, Quarterly Review, and Records Update works on a locked staged database and atomically publishes the result after validation. Annual Report writes an HTML dashboard, isolate and genome ledgers, workflow history, and the latest decision-change audit.

Office-stage reports are named onboarding_report.tsv, onboarding_summary.json, status_meeting_report.tsv, records_update_report.tsv, annual_report.html, it_desk_report.tsv, and it_desk_summary.json.

Before a production run:

branchmanager it-desk --db project.sqlite --ref gtdb_ssu_reps.fna.gz \
  --tree-method fasttree --strict -o it_desk

What BranchManager does

BranchManager is aimed at ranking marker-gene sequence evidence for isolate follow-up. It helps identify lineages that may have been missed because of primer bias, sparse reference coverage, taxonomy lag, or conservative filtering. Many targets are not new in nature — they are often new to the reference record or underrepresented in existing collections.

That is why BranchManager combines reference-aware classification with novelty scoring, Sanger QC, baseline/reference comparisons, partner metadata, and warning layers, rather than treating every long branch as a discovery.