Annostat

Annostat analyzes bacterial genome annotations from matching GFF3 and FASTA files. Local analysis uses only the Python standard library and supports multi-record FASTA files.

Features

• Counts CDS, RNA types, all feature types, hypothetical proteins, gene annotations, and CDS with COG categories
• Exports a compact feature overview as CSV or TSV
• Extracts every CDS on the correct strand as nucleotide and amino-acid multi-FASTA
• Calculates codon usage percentages, start-codon counts, and individual COG category counts
• Reports GC content, coding density, CDS overlaps and containment, pseudogenes, partial features, structural-RNA completeness, frame issues, and ambiguous bases
• Supports optional CDS exports filtered by length, COG annotation, and hypothetical-protein status without changing the complete analysis
• Builds a self-contained offline HTML report with provenance and performance timings
• Saves three focused, dependency-free SVG charts for COG categories, CDS lengths, and table-aware start codons; the COG chart is omitted when the source has no COG data
• Streams CDS processing and FASTA generation to reduce runtime and peak memory
• Compares two or more local or NCBI-hosted annotations using normalized genome, annotation-completeness, COG, start-codon, and QC profiles
• Reports pairwise Jensen-Shannon profile distances and exact gene-symbol/COG-ID overlap without presenting annotation labels as inferred orthologs
• Downloads versioned RefSeq or GenBank assemblies through the official NCBI Datasets command-line tool when network connectivity is requested
• Separates deterministic FASTA/GFF3 validation from biological inspection, with stable rule IDs, severity levels, schema/ruleset versions, and source citations
• Records SHA-256 hashes for reproducibility and writes canonical JSON plus TSV validation results suitable for CI pipelines
• Correctly joins repeated-ID multipart CDS features in biological order on both strands and validates phase continuity before translation
• Supports NCBI prokaryotic translation tables 4, 11, and 25, automatically selecting a consistent CDS or region transl_table declaration when present
• Applies position-specific NCBI transl_except annotations for selenocysteine, pyrrolysine, explicitly annotated termination codons, and OTHER residues represented conservatively as X
• Aggregates completed inspections into a script-free cohort HTML report, normalized TSV matrix, and machine-readable JSON

Required and optional components

No third-party Python package is required for local analysis, report generation, tables, FASTA exports, or SVG plots.

Core functionality

Annostat provides the following core analysis outputs:

• Counts CDS, RNA features by type, hypothetical CDS, CDS with gene names, CDS with COG categories, and other feature types.
• Writes a CSV or TSV feature overview containing ID, type, start, stop, gene, product, and strand, with missing values left empty.
• Exports all CDS as nucleotide and translated amino-acid multi-FASTA files.
• Calculates pooled codon usage percentages, start-codon counts, and detailed COG-category statistics, including multi-category assignments.
• Generates at least two meaningful scientific plots; a normal run produces CDS length, start-codon, and—when available—COG-category SVG figures.
• Provides an argparse command-line interface and modular Python structure with documented functions, classes, and methods.

Filtering, quality control, HTML reports, performance profiling, comparative analysis, cohort summaries, and NCBI connectivity extend the core workflow.

Input assumptions and scope

• GFF3 records must contain nine tab-separated fields with 1-based inclusive coordinates. The FASTA identifier is the first whitespace-delimited header token.
• CDS strand and phase are honored. Circular-origin CDS coordinates extending past the sequence length are supported when a region feature declares Is_circular=true.
• Repeated CDS IDs are interpreted as one multipart biological feature. Segments must use the same sequence ID, strand, and Parent; phase continuity is checked, and segments are joined in biological order before translation.
• COG coverage describes categories actually present in the annotation source. An annotation with no COG category fields is reported as unavailable, not as 0%.
• Annostat uses the consistent CDS or region transl_table declared in GFF3. When the annotation declares no table, it falls back to bacterial table 11. --genetic-code is an explicit override for annotations without that metadata; an override that conflicts with a declared table is rejected instead of silently mistranslating CDS features. Annostat never guesses a genetic code from sequence composition.

Installation

Annostat requires Python 3.10 or newer and has no third-party runtime dependencies. Clone the repository and create an isolated environment.

Linux or WSL:

git clone https://github.com/Karudhoru/annostat.git
cd annostat
python3 -m venv .venv
source .venv/bin/activate

Windows PowerShell:

git clone https://github.com/Karudhoru/annostat.git
Set-Location annostat
py -3 -m venv .venv
.\.venv\Scripts\Activate.ps1

Install the current release from the checkout and verify the executable:

python -m pip install .
annostat --version
annostat --help

Expected version output:

Annostat 1.0.0

Contributors should use python -m pip install -e . so source changes take effect without reinstalling. If the executable is unavailable, the equivalent module form is python -m annostat.

Quick-start guide

1. Check the inputs

Confirm that you have one genome FASTA and its matching GFF3 annotation. Sequence IDs must agree—for example, a FASTA record named chromosome must be referenced as chromosome in the first GFF3 column.

2. Validate the paired files

Run deterministic integrity checks before biological interpretation:

annostat validate \
  -f data/GCF_000007145.1.fna \
  -g data/GCF_000007145.1.gff3 \
  -o validation

Example output for the included dataset:

Annostat 1.0.0 | annotation validation PASS
  Errors                   0
  Warnings                 0
  Findings                 0
  Files written            2

The command writes validation.json and validation.tsv. Exit status 0 means no structural errors were found; status 1 means the configured threshold was reached. Use --fail-on warning for strict CI, or --fail-on never when the report should never stop a pipeline. A validation pass means the paired files are internally interpretable; it does not prove that every biological annotation is correct.

3. Inspect one annotation

annostat inspect \
  -f data/GCF_000007145.1.fna \
  -g data/GCF_000007145.1.gff3 \
  -o results \
  --table-format tsv

Here -f selects the required FASTA, -g selects the required GFF3, and -o chooses the output directory. Add --table-format tsv when a TSV overview is preferred over CSV. The original command without the inspect word remains a backward-compatible alias. Translation-table selection is automatic when the GFF3 CDS or region rows contain transl_table; otherwise table 11 is used. The included example produces a 5,076,188 bp analysis with 4,295 CDS. For a local annotation without translation-table metadata, select a supported table explicitly with --genetic-code 4, --genetic-code 11, or --genetic-code 25.

4. Review the result package

Open results/report.html in a browser first. Then use:

• summary.json for the complete machine-readable summary;
• tables/ for feature, codon, COG, start-codon, and QC data;
• sequences/ for nucleotide and protein CDS FASTA files; and
• plots/ for editable SVG figures.

5. Summarize a cohort

After inspecting several genomes into separate result directories, list those directories explicitly or pass their common parent directory:

annostat summarize results/sample-a results/sample-b -o cohort

The output contains cohort.html, cohort.tsv, and cohort.json. Samples are ordered deterministically. Missing source-dependent values such as COG coverage remain NA/null, never biological zero. Multiple source Annostat versions are retained and flagged in the report.

Common workflows

The following sections show filtered exports, comparative analysis, NCBI references, performance profiling, and complete output descriptions.

Without installing, module execution remains available from the repository root: python -m annostat --help.

Use --table-format tsv for a tab-separated feature overview. Add --profile to measure peak Python memory and print per-stage timings.

Filtered CDS exports

Filters create an additional filtered/ directory while retaining every normal analysis file. Criteria are combined, so a CDS must satisfy all enabled options:

annostat \
  -f genome.fna \
  -g annotations.gff3 \
  -o results \
  --min-cds-length 300 \
  --require-cog \
  --exclude-hypothetical

Available criteria are --min-cds-length, --max-cds-length, --require-cog, and --exclude-hypothetical.

Comparative analysis

Compare two or more labelled local genomes with repeatable --genome options:

annostat compare \
  --genome genome-a data/GCF_000007145.1.fna data/GCF_000007145.1.gff3 \
  --genome genome-b data/GCF_001050915.2.fna data/GCF_001050915.2.gff3 \
  --output comparison

Each genome is structurally validated and independently uses its declared transl_table or the table-11 fallback. For an annotation without transl_table, a per-genome override uses the same label supplied to --genome (or the reference accession):

annostat compare \
  --genome sr1 sr1.fna sr1.gff3 \
  --genome ecoli ecoli.fna ecoli.gff3 \
  --genetic-code sr1 25 \
  --output comparison

The comparison package is intentionally compact:

comparison/
|-- comparison.html
|-- comparison.json
|-- tables/
|   |-- dataset_metrics.tsv
|   `-- pairwise_comparisons.tsv
`-- plots/
    |-- dataset_overview.svg
    `-- cog_comparison.svg (only when every input supplies COG categories)

dataset_metrics.tsv provides raw genome/feature counts together with GC, coding density, CDS per Mb, annotation coverage, start-codon usage, and QC warnings. pairwise_comparisons.tsv contains metric deltas, percentage deltas, Jensen-Shannon distances for COG and start-codon profiles, and Jaccard overlap for exact gene symbols and explicit COG identifiers.

For two datasets, the COG figure shows percentage-point differences. For three or more datasets, it becomes a dataset-by-category heatmap. Exact gene-symbol overlap describes annotation labels only; it is not orthology, ANI, phylogeny, or a core/accessory pangenome analysis.

Compare against external NCBI data

NCBI access is optional and uses NCBI's maintained datasets executable. Follow the official installation instructions, then add an external reference directly to a local comparison:

annostat compare \
  --genome local-sample sample.fna sample.gff3 \
  --reference GCF_000007145.1 \
  --output comparison

--reference is repeatable, so a comparison may contain only external assemblies or any mixture of local and NCBI inputs. Annostat downloads genome FASTA, GFF3, and assembly metadata under comparison/external_inputs/, then runs every input through the same comparison engine:

annostat compare \
  --reference GCF_000007145.1 \
  --reference GCF_001050915.2 \
  --output comparison

The report identifies organisms and whether each pair belongs to the same or different species when the input metadata makes that determination possible. It never assumes that labelled datasets are strains. NCBI assembly names, taxonomy identifiers, annotation providers, and release metadata are retained in comparison.json. Versioned GCF/GCA accessions in the HTML provenance table link to their NCBI Datasets assembly pages.

COG results are source-dependent. Bakta annotations commonly include COG categories, while current NCBI RefSeq GFF3 packages may not. When a source does not supply COG fields, Annostat reports COG coverage and pairwise COG statistics as Not available and omits the COG comparison plot; missing data is never presented as biological zero coverage.

For users who only need the source files, the standalone downloader remains available:

annostat fetch GCF_000007145.1 GCF_001050915.2 --output ncbi_data

The NCBI_API_KEY environment variable is honored by the official client. Local analysis never requires a network connection, and Annostat does not transmit local FASTA or GFF3 files.

The output directory contains:

results/
├── report.html
├── summary.json
├── tables/
│   ├── features.csv (or features.tsv)
│   ├── codon_usage.csv
│   ├── start_codons.csv
│   ├── cog_categories.csv
│   └── annotation_issues.csv
├── validation/
│   ├── validation.json
│   └── validation.tsv
├── sequences/
│   ├── cds_nucleotide.fasta
│   └── cds_protein.fasta
└── plots/
    ├── cog_categories.svg (when COG categories are available)
    ├── cds_lengths.svg
    └── start_codons.svg

When filters are active, filtered/ additionally contains the selected feature table and paired nucleotide/protein FASTA files.

report.html works without a server or internet connection and embeds the SVG figures directly. The separate SVG files remain available for editing and use in documents. Codon usage remains available as the required percentage table; a dense 64-codon heatmap is intentionally not generated by default.

annotation_issues.csv is a review aid rather than an automatic pass/fail verdict. Short CDS overlaps are common in compact bacterial genomes, so ordinary overlaps and adjacent duplicate labels are informational. Containment, incomplete frames, missing structural RNA classes, internal stops, and ambiguous bases are reported as warnings that merit inspection in their biological context.

Validation versus inspection

annostat validate answers whether the files can be interpreted consistently: GFF3 syntax, FASTA symbols, matching sequence IDs, coordinate bounds, Parent/ID relationships, multipart-feature consistency, CDS phase continuity, and valid, consistent CDS translation-table declarations. These checks are deterministic and may safely control a CI exit status.

annostat inspect first requires structural validation to pass, then reports biological review candidates such as internal stops, missing terminal codons, ambiguous CDS bases, overlaps, partial features, pseudogenes, duplicated adjacent labels, and structural-RNA coverage. These findings are context-dependent and do not automatically make an annotation invalid. Boundary-specific partial markers suppress only the affected start- or stop-codon assumption. Explicit pseudo, transl_except, and exception annotations suppress checks whose assumptions do not apply.

Every validation result embeds its rule definitions, rationale, source URL, ruleset version, schema version, Annostat version, and input SHA-256 hashes. JSON and TSV findings use canonical ordering and contain no timestamp, so repeated runs on identical paths and bytes produce byte-identical validation artifacts. A scientific_fingerprint in both validation and inspection JSON excludes file paths and performance timings, allowing scientific results to be compared across machines and output directories.

Testing and verification

Run compilation and all automated tests with the interpreter used to install Annostat:

python -m compileall -q annostat tests
python -m unittest discover -v

For a final end-to-end check, analyze one supplied genome and confirm that results/report.html, results/summary.json, the tables, FASTA exports, and SVG plots are created:

annostat inspect \
  -f data/GCF_000007145.1.fna \
  -g data/GCF_000007145.1.gff3 \
  -o results \
  --table-format tsv

Example run

annostat inspect \
  -f data/GCF_000007145.1.fna \
  -g data/GCF_000007145.1.gff3 \
  -o results \
  --table-format tsv

Example output:

Annostat 1.0.0 | bacterial genome annotation analysis
FASTA: /path/to/annostat/data/GCF_000007145.1.fna
GFF3:  /path/to/annostat/data/GCF_000007145.1.gff3

[1/5] Reading GFF3 annotations and FASTA sequences
[2/5] Streaming, translating, and writing CDS sequences
[3/5] Calculating statistics and writing analysis tables
[4/5] Rendering scientific visualizations
[5/5] Building the offline HTML report

Analysis summary
--------------------------------------------------------
  Genome size                              5,076,188 bp
  Sequences                              1 (1 circular)
  Features                                        4,498
  CDS                                             4,295
  RNA features                                      181
  Genome GC                                      65.07%
  Coding density                                 85.33%
  Hypothetical CDS                         619 (14.41%)
  COG-annotated CDS                      2,881 (67.08%)
  Recognized starts                      4,294 (99.98%)
  QC review                1 warning, 872 informational findings
--------------------------------------------------------
  Output                   /path/to/annostat/results
  Files written            14
  Completed in             0.56 seconds

Runtime depends on the input size and system. The supplied bacterial genome typically completes in under a second on a modern desktop. Use --profile when you need measured stage timings and peak-memory information; profiling itself adds overhead and should not be used for speed benchmarks.

Scientific basis

The QC checks follow the internal-consistency emphasis of the NCBI Prokaryotic Genome Annotation Standards and the attribute conventions in the NCBI GFF3 documentation. FASTA validation accepts the NCBI/IUPAC nucleotide alphabet, and translation tables and initiator codons follow the NCBI Genetic Codes. The report treats overlaps as review candidates because overlapping genes are common in compact bacterial genomes. Translation uses the bacterial code and alternative-start behavior described by Prodigal. The project remains an annotation-analysis tool rather than a replacement for comprehensive annotation pipelines such as Bakta or Prokka.

Comparative profiles are normalized to avoid treating larger genomes as automatically enriched. Full ortholog and pangenome inference remains the domain of tools such as Panaroo, while sequence-level average nucleotide identity should be calculated by a specialized implementation such as FastANI.