alpha_fold_viewer

AF3 ZIP → Standalone HTML Report — a single-file tool that converts AlphaFold3 output ZIP files into beautiful, self-contained HTML reports.

What it does

Takes an AlphaFold3 prediction ZIP file and produces a single HTML file containing:

• Input summary — sequences in FASTA format with copy buttons, chain types, lengths
• Confidence overview — all models ranked by ranking_score with ipTM, pTM, fraction disordered, clash status
• Chain sequences — full sequences with per-residue pLDDT coloring and interface residue highlighting
• Sequence heatmap strips — linear pLDDT bars with interface position markers
• PAE heatmaps — full predicted aligned error matrices with chain boundaries (embedded as base64 images)
• Interface analysis — inter-chain contacts with residue counts, mean PAE, pLDDT, and high-confidence percentages
• Per-model details — collapsible sections with chain info and interface residue ranges

The HTML is fully standalone — all images are embedded as data URIs, CSS is inline, no external dependencies. Open it in any browser, share via email, or include in presentations.

Works with any AlphaFold3 output: protein homodimers, heterodimers, protein+DNA complexes, multi-chain assemblies.

PAE Heatmap (CTCF protein + DNA complex, 3 chains)

Interface PAE Sub-matrix

Sequence Strip with pLDDT and Interface Markers

Colored Sequence with Interface Residues

Installation

From PyPI

pip install alpha-fold-viewer

From source

git clone https://github.com/aglabx/alpha_fold_viewer.git
cd alpha_fold_viewer
pip install .

Requirements: Python 3.8+, numpy, scipy, matplotlib.

Usage

# Basic usage — generates fold_ctcf_report.html
af3-report fold_ctcf_dimer.zip

# Custom output path
af3-report fold_ctcf_dimer.zip -o ctcf_report.html

# Stricter contact distance (default: 8.0 Å)
af3-report fold_ctcf_dimer.zip --contact-dist 6.0

# Keep extracted temp files for debugging
af3-report fold_ctcf_dimer.zip --keep-tmp

# Also works as a Python script
python af3_report.py fold_ctcf_dimer.zip

CLI Reference

af3-report INPUT_ZIP [-o OUTPUT_HTML] [--contact-dist 8.0] [--keep-tmp]

Positional:
  INPUT_ZIP          Path to AlphaFold3 output ZIP file

Options:
  -o, --output       Output HTML file (default: {zip_name}_report.html)
  --contact-dist     Inter-atomic contact threshold in Å (default: 8.0)
  --keep-tmp         Keep extracted temporary files

Output Description

Confidence Overview

Models are sorted by ranking_score (highest first). The best model is highlighted in green. Columns:

Column	Description
Ranking Score	AF3 composite confidence metric (higher = better)
ipTM	Interface predicted TM-score (0–1, higher = better interface)
pTM	Predicted TM-score for overall structure
Frac. Disordered	Fraction of residues predicted as disordered
Clash	Whether the model has steric clashes

Chain Sequences

Each chain is displayed with:

• Full sequence colored by per-residue pLDDT (green ≥90, cyan ≥70, yellow ≥50, red <50)
• Interface residues highlighted with cyan background
• Linear heatmap strip showing pLDDT along the sequence with interface markers

Interface Analysis

For multi-chain models, inter-chain contacts are detected using a KDTree spatial search. Each interface reports:

Metric	Description
Res. A / Res. B	Number of residues at the interface per chain
Atom Contacts	Total inter-chain atom pairs within contact distance
pLDDT A / pLDDT B	Mean predicted local confidence at interface residues
Avg PAE	Mean predicted aligned error across interface residue pairs
PAE <10Å	Percentage of PAE values below 10Å (higher = more confident)
High-conf	Percentage of contacts where both atoms have pLDDT ≥ 70

PAE Heatmaps

Each model gets a full PAE matrix heatmap with chain boundary lines. The colormap runs from dark blue (low PAE = high confidence) through green/yellow to red (high PAE = low confidence). Scale: 0–30 Å.

For multi-chain models, per-interface sub-matrices are also shown with mean PAE and <10Å percentage annotations.

How it works

1. Extracts the AF3 ZIP to a temporary directory
2. Auto-discovers model files (*_model_*.cif, *_full_data_*.json, *_summary_confidences_*.json)
3. Parses mmCIF structures to extract atom coordinates, chain IDs, pLDDT values
4. Loads PAE matrices from full_data JSONs
5. Detects inter-chain interfaces using scipy KDTree
6. Cross-references interfaces with PAE data
7. Generates PAE heatmaps in-memory using matplotlib (→ base64 PNGs)
8. Assembles everything into a single standalone HTML file

License

MIT