Adjudicator

Version: 1.0.0 · License: Apache-2.0 · Status: Production/Stable
PyPI: pip install adjudicator · Source: ncgr/Adjudicator
Author: Connor Cameron · ctc@ncgr.org · National Center for Genome Resources

---

Overview

Adjudicator is a command-line tool for collapsing and filtering structural genome annotations across multiple sources. It uses best-hit HMM domain scores from gene family assignments (via the Legume Information System) to compare overlapping gene models and select the best-supported annotation for all overlapping models.

Two commands are provided:

• collapse — Merge overlapping gene models from two or more annotators, selecting the best model per region.
• repeat-filter — Remove gene models that overlap known repeat or transposon regions beyond a configurable coverage threshold.

---

Requirements

Requirement	Version
Python	≥ 3.10
click	≥ 8.1
intervaltree	≥ 3.2.1
sortedcontainers	≥ 2.4.0

---

Installation

pip install adjudicator

adjudicator --version
# adjudicator, version 1.0.0

---

Input File Formats

TSV Sample Sheet (--input-tsv)

Tab-separated. Lines beginning with # and blank lines are skipped.

Column	Type	Description
1	string	Unique label for this evidence set.
2	path	Path to the .gff3 structural annotation file.
3	path	Path to the .gfa LIS gene family assignment file.

Row order determines precedence when gene models have equivalent scores.

# label	gff3_path	gfa_path
maker	/data/ann/maker.gff3	/data/fam/maker.gfa
helixer	/data/ann/helixer.gff3	/data/fam/helixer.gfa
stringtie	/data/ann/stringtie.gff3	/data/fam/stringtie.gfa

GFF3 (.gff3)

Standard GFF3 format with a three-level hierarchy: gene → mRNA → exon. See the GFF3 specification.

GFA (.gfa)

LIS gene family assignment files produced by the Legume Information System gene family pipeline.

---

Commands

collapse

Collapses overlapping structural annotations across all entries in the TSV. Processing is hierarchical: Row 1 vs. Row 2 produces an intermediate result, which is then compared against Row 3, and so on.

Synopsis

adjudicator collapse --input-tsv <FILE> [OPTIONS]

Options

Option	Short	Type	Default	Valid range	Description
--input-tsv	-i	path	(required)	—	Tab-separated sample sheet.
--min-overlap	-m	float	0.00001	0.0 – 1.0	Minimum fractional overlap of feature A by feature B to consider them overlapping.
--no-orphans	-n	flag	False	—	Exclude genes with no gene family assignment from the output.
--output-dir	-o	path	.	—	Directory to write output files. Created if it does not exist.
--strict / --no-strict		flag	False	—	Exit with error if any referenced input file does not exist on disk.
--verbose	-v	flag	False	—	Print per-sample file paths and processing steps to stdout.

Output Files

<output-dir>/
├── A_B.wao.gff3                       # Overlap intersections
├── A_B.unique_b.gff3                  # Gene models unique to annotator B
├── A_B.final.gff3                     # Adjudicated gene IDs
├── A_B.gfa                            # Merged gene family assignments
└── A_B.final.wsubfeatures.gff3        # ✅ Primary output

Examples

adjudicator collapse \
    --input-tsv samples.tsv \
    --output-dir results/collapse/

adjudicator collapse \
    --input-tsv samples.tsv \
    --no-orphans \
    --min-overlap 0.4 \
    --output-dir results/collapse/ \
    --verbose

---

repeat-filter

Filters gene models from each entry in the TSV against a reference repeat annotation. Gene models whose exons exceed --max-coverage overlap with a repeat region are removed.

Synopsis

adjudicator repeat-filter --input-tsv <FILE> --annotation <FILE> [OPTIONS]

Options

Option	Short	Type	Default	Valid range	Description
--input-tsv	-i	path	(required)	—	Tab-separated sample sheet.
--annotation	-a	path	(required)	—	GFF3 file of repeat regions to filter against.
--max-coverage	-m	float	0.4	0.0 – 1.0	Maximum fractional overlap between a gene's exons and a repeat region before the model is removed.
--output-dir	-o	path	.	—	Directory to write output files. Created if it does not exist.
--strict / --no-strict		flag	False	—	Exit with error if any referenced input file does not exist on disk.
--verbose	-v	flag	False	—	Print per-sample file paths and processing steps to stdout.

Output Files

<output-dir>/
├── <label>_repeat_filter.wao.gff3                  # Overlap intersections
└── <label>_repeat_filter.final.wsubfeatures.gff3   # ✅ Primary output

Examples

adjudicator repeat-filter \
    --input-tsv samples.tsv \
    --annotation repeats.gff3 \
    --output-dir results/filtered/

adjudicator repeat-filter \
    --input-tsv samples.tsv \
    --annotation transposons.gff3 \
    --max-coverage 0.3 \
    --output-dir results/filtered/ \
    --strict \
    --verbose

---

Workflow

# Step 1: Filter repeat regions
adjudicator repeat-filter \
    --input-tsv raw_samples.tsv \
    --annotation repeats.gff3 \
    --output-dir step1_filtered/

# Step 2: Rewrite TSV to point to filtered outputs (GFA paths unchanged)
awk -F'\t' 'OFS="\t" { $2="step1_filtered/"$1"_repeat_filter.final.wsubfeatures.gff3"; print }' \
    raw_samples.tsv > filtered_samples.tsv

# Step 3: Collapse filtered annotations
adjudicator collapse \
    --input-tsv filtered_samples.tsv \
    --output-dir step2_collapsed/

---

Error Reference

Condition	--strict off	--strict on
Input file not found	Warning to stderr	Error: The following files were not found: ...
TSV row has wrong column count	BadParameter: Line N: expected 3 columns, got N.	Same
Label (column 1) is empty	BadParameter: Line N: column 1 (label) must not be empty.	Same
GFF3 path does not end in .gff3	BadParameter: Line N: column 2 must end in '.gff3'	Same
GFA path does not end in .gfa	BadParameter: Line N: column 3 must end in '.gfa'	Same
TSV contains no data rows	Error: No data rows found in '<file>'.	Same

---

Glossary

Term	Definition
Gene model	A predicted gene structure represented as a gene → mRNA → exon hierarchy in GFF3.
GFF3	Generic Feature Format version 3. Tab-delimited format for genomic features and their hierarchical relationships.
GFA	Gene Family Assignment file from the LIS pipeline, containing HMM domain scores used to rank competing gene models.
Adjudication	Selection of one gene model from a set of overlapping candidates based on HMM score evidence.
Orphan gene	A gene model with no gene family assignment in the GFA file.
WAO intersection	A bedtools-style "write all overlaps" operation reporting fractional overlap between features across two GFF3 files.