gel-annotation - SKILL.md Agent Skill

name: gel-annotation description: Annotate SDS-PAGE gel images with molecular weight markers and lane labels. Use when asked to label, annotate, or analyze gel photos. Detects lanes and bands automatically via intensity profiling (scikit-image), assigns MW from known protein ladders, and outputs SVG-quality annotated images.

Gel Annotation

Automated SDS-PAGE gel annotation: band detection, MW assignment, lane labeling.

Quick Start

# Basic annotation with PageRuler 26616 ladder (default)
python3 scripts/annotate_gel.py input.png -o annotated.png

# Specify ladder and ladder lane
python3 scripts/annotate_gel.py input.png --ladder pageruler-26616 --ladder-lane 0

# Custom MW values
python3 scripts/annotate_gel.py input.png --mw 140,115,80,70,50,40,30,25,15,10

# Custom lane labels (comma-separated; use "Ladder" for the marker lane)
python3 scripts/annotate_gel.py input.png --labels "Ladder,Input,FT,Wash,Elution"

# Output SVG (vector labels, raster gel)
python3 scripts/annotate_gel.py input.png --format svg -o annotated.svg

# High-res PNG from SVG (2x default)
python3 scripts/annotate_gel.py input.png --format png --scale 2 -o annotated_hires.png

CLI Reference

Located at skills/gel-annotation/scripts/annotate_gel.py.

Flag	Default	Description
`input`	(required)	Input gel image (PNG, JPEG, HEIC via sips conversion)
`-o, --output`	`annotated.png`	Output file path
`--ladder`	`pageruler-26616`	Preset ladder name
`--mw`	(from ladder)	Comma-separated MW values (kDa), overrides ladder
`--ladder-lane`	`0`	Lane index for the MW ladder (0 = leftmost)
`--labels`	auto (Ladder,A,B,...)	Comma-separated lane labels
`--format`	`png`	Output format: `png`, `svg`
`--scale`	`2`	PNG render scale (via rsvg-convert)
`--ref-bands`	`70,40,25`	Reference bands to highlight in red
`--band-sigma`	`4`	Gaussian sigma for band detection smoothing
`--lane-sigma`	`8`	Gaussian sigma for lane detection smoothing
`--min-prominence`	`3`	Minimum peak prominence for band detection
`--font-size`	`16`	Base font size for labels
`--no-lines`	false	Skip dashed MW indicator lines

Supported Ladders

Name	Bands (kDa)	Catalog
`pageruler-26616`	140, 115, 80, 70, 50, 40, 30, 25, 15, 10	Thermo 26616
`pageruler-26619`	250, 130, 100, 70, 55, 35, 25, 15, 10	Thermo 26619

Add new ladders in the LADDERS dict in the script.

Pipeline

Convert HEIC → PNG via sips (macOS) if needed
Detect lanes — vertical intensity profile aggregated across gel height, peak detection
Detect bands — per-lane horizontal intensity profile, Gaussian-smoothed peak detection
Assign MW — match detected bands in ladder lane to known MW values (count must match)
Render SVG — gel as embedded raster <image>, all labels as vector <text> elements
Export — SVG direct or PNG via rsvg-convert at specified scale

Tuning

If band detection misses bands or picks up noise:

Adjust --band-sigma (higher = smoother, fewer peaks)
Adjust --min-prominence (higher = only strong bands)
For overloaded lanes, increase prominence threshold

If lane detection fails:

Adjust --lane-sigma
Manually specify lane count or positions (future flag)

Dependencies

Python 3: numpy, scipy, scikit-image, Pillow
rsvg-convert (for PNG export from SVG; install via brew install librsvg)
sips (macOS built-in, for HEIC conversion)