Prediction Alignment Examples

This page demonstrates how prediction alignment works for different variant types using simulated example data. Each example shows both 1D predictions (e.g., chromatin accessibility, transcription factor binding) and 2D predictions (e.g., Hi-C contact maps).

Variant Type Examples

INS (Insertion)

Insert NaN bins in reference where insertion occurs in alternate.

1D Example

INS 1D Alignment

2D Example

INS 2D Alignment

DEL (Deletion)

Insert NaN bins in ALT where deletion removes sequence from reference.

1D Example

DEL 1D Alignment

2D Example

DEL 2D Alignment

DUP (Duplication)

Same as insertion - duplications add sequence like insertions.

1D Example

DUP 1D Alignment

2D Example

DUP 2D Alignment

INV (Inversion)

Mask inverted bins in both REF and ALT. This follows the original Supremo implementation.

1D Example

INV 1D Alignment

2D Example

INV 2D Alignment

BND (Breakend)

Breakends join two distant genomic loci, creating fusion sequences. BND alignment requires special handling with dual loci.

Alignment strategy:

Reference: Generate predictions from two separate loci (left and right breakpoints), then concatenate
Alternate: Single prediction from the fused sequence
1D: Concatenate left + right reference
2D: Assemble chimeric matrix from quadrants

1D Example

BND 1D Alignment

2D Example

BND 2D Alignment

Additional Resources

Prediction Alignment User Guide - Complete API reference
03_prediction_alignment.ipynb - Notebook example
Variant Classification - Variant type decision tree

Reproducing These Examples

To regenerate these visualizations:

# Run the example generation script
poetry run python create_prediction_alignment_examples.py

# Images will be saved to:
# docs/_static/images/prediction_alignment_examples/

The script source: create_prediction_alignment_examples.py