Orientation-aware Fragmentation (OCF)

Command: krewlyzer ocf

Purpose

Computes orientation-aware cfDNA fragmentation (OCF) values in tissue-specific open chromatin regions. Enables tissue-of-origin analysis from cfDNA.

Processing Flowchart

flowchart LR
    BED["sample.bed.gz"] --> RUST["Rust Pipeline"]
    OCR["Open Chromatin Regions"] --> RUST
    GC["GC Correction"] --> RUST

    RUST --> OCF["OCF.tsv"]
    RUST --> SYNC["OCF.sync.tsv"]

    subgraph "With --pon-model"
        OCF --> PON["z-score normalization"]
    end

    subgraph "With --target-regions"
        RUST --> OCF_ON["OCF.ontarget.tsv"]
    end

Python/Rust Architecture

flowchart TB
    subgraph "Python (CLI)"
        CLI["ocf.py"] --> UP["unified_processor.py"]
        UP --> ASSETS["AssetManager"]
    end

    subgraph "Rust Backend"
        UP --> RUST["_core.run_unified_pipeline()"]
        RUST --> GC["GC correction"]
        GC --> OCF_CALC["OCF strand counting"]
    end

    subgraph "Python (Post-processing)"
        OCF_CALC --> PROC["Output cleanup"]
        PROC --> PON["PON z-scores"]
        PON --> OUT["OCF.tsv"]
    end

Biological Context

OCF (Sun et al., 2019) measures the phasing of upstream (U) and downstream (D) fragment ends in open chromatin regions, informing tissue-of-origin of cfDNA.

Usage

# Basic usage
krewlyzer ocf -i sample.bed.gz -o output_dir/ --genome hg19

# With PON for z-scores
krewlyzer ocf -i sample.bed.gz -o output/ -P tissue.pon.parquet

# Panel data with on/off-target split
krewlyzer ocf -i sample.bed.gz -o output/ \
    --target-regions MSK-ACCESS_targets.bed

CLI Options

Output Files

Formulas

OCF Score Calculation

Where: - \(\text{Right}_{-60}\) = Right fragment ends at -60bp from OCR center (phased) - \(\text{Left}_{+60}\) = Left fragment ends at +60bp from OCR center (phased) - \(\text{Left}_{-60}\), \(\text{Right}_{+60}\) = Background (unphased)

Calculation Details: 1. Fragments are mapped relative to the center of the Open Chromatin Region (OCR) 2. Left/Right ends counted in 10bp bins across ±1000bp window 3. Counts normalized by total sequencing depth

PON Normalization

When --pon-model is provided, OCF output includes z-score columns:

Output Columns with PON

Z-Score Interpretation

Panel Mode

For targeted sequencing panels (MSK-ACCESS):

krewlyzer ocf -i sample.bed.gz -o output/ \
    --target-regions MSK-ACCESS_targets.bed

How Panel OCF Works

In panel mode, OCF produces two complementary outputs using a sophisticated two-pass approach:

flowchart TB
    subgraph "Pass 1: Genome-Wide"
        OCR1["All 50K OCR regions"] --> RUN1["OCF analysis"]
        FRAGS1["All fragments"] --> RUN1
        RUN1 --> OCF["OCF.tsv"]
        RUN1 --> SYNC["OCF.sync.tsv"]
    end

    subgraph "Pass 2: Panel-Focused"
        OCR2["Panel OCRs (~500)"] --> RUN2["OCF analysis"]
        TARGET["Target regions"] --> FILTER["Filter OCRs"]
        OCR1 --> FILTER
        FILTER --> OCR2
        FRAGS2["On-target frags"] --> RUN2
        RUN2 --> OCFON["OCF.ontarget.tsv"]
    end

Panel OCF Regions

Before the ontarget OCF run, the genome-wide OCR atlas (~50,000 regions) is filtered to keep only regions that overlap with panel targets (+2kb promoter extension). For a typical panel like MSK-ACCESS:

Output Files

Why Both Filters?

flowchart LR
    subgraph "On-Target Fragments"
        CAP["Captured near panel genes"]
    end

    subgraph "Panel OCR Regions"
        POCR["OCRs near panel genes"]
    end

    CAP --> BOTH["Same genomic space"]
    POCR --> BOTH
    BOTH --> SIGNAL["Maximum signal-to-noise"]

On-target fragments and panel OCR regions both focus on the same genomic space (near panel target genes), so combining both filters maximizes the signal-to-noise ratio for tissue-of-origin detection.

Example: MSK-ACCESS Panel

Clinical Interpretation

Healthy Plasma Baseline

Cancer-Specific Patterns

Interpretation Guide

See Also

• Input File Formats - Region BED format for --ocr-input
• PON Models – Tissue baseline models
• Citation – Sun et al. paper
• Troubleshooting – hg38 issues