Skip to content

MSK-ACCESS Quickstart

This guide walks through analyzing MSK-ACCESS panel data with Krewlyzer.

Prerequisites

  • Krewlyzer installed (pip install krewlyzer or Docker)
  • MSK-ACCESS BAM file (sorted, indexed)
  • Reference genome (hg19.fa)

Basic Panel Analysis

krewlyzer run-all -i ACCESS_sample.bam \
    --reference hg19.fa \
    --output results/ \
    --assay xs2 \
    --target-regions MSK-ACCESS-v2_targets.bed

What This Does

Flag Effect
--assay xs2 Enables MSK-ACCESS v2 optimizations
--target-regions Splits outputs into on/off-target

Output Files

Standard Outputs

File Description
sample.bed.gz Extracted fragments
sample.FSC.tsv Fragment size coverage
sample.FSC.ontarget.tsv FSC for on-target regions
sample.FSR.tsv Fragment size ratios
sample.FSD.tsv Size distribution by arm
sample.WPS.parquet Nucleosome protection (genome-wide)
sample.OCF.tsv Tissue of origin footprint
sample.EndMotif.tsv End motif frequencies
sample.MDS.tsv Motif Diversity Score

Panel-Specific Outputs (with --assay)

File Description
sample.FSC.gene.tsv Gene-level FSC (146 genes for xs2)
sample.WPS.panel.parquet Panel-focused WPS (~2k anchors)

With PON Normalization

For z-score computation against healthy baselines:

krewlyzer run-all -i ACCESS_sample.bam \
    --reference hg19.fa \
    --output results/ \
    --assay xs2 \
    --target-regions MSK-ACCESS-v2_targets.bed \
    --pon-model xs2.pon.parquet

This adds z-score columns to all outputs (e.g., z_core_short, wps_nuc_z, mds_z).

JSON Output for ML

Generate a unified JSON file for machine learning pipelines:

krewlyzer run-all -i ACCESS_sample.bam \
    --reference hg19.fa \
    --output results/ \
    --assay xs2 \
    --target-regions MSK-ACCESS-v2_targets.bed \
    --pon-model xs2.pon.parquet \
    --generate-json

This creates sample.features.json with all features. See JSON Output for schema details.

Assay Options

Assay Code Panel Genes WPS Anchors
xs1 MSK-ACCESS v1 128 1,611
xs2 MSK-ACCESS v2 146 1,820

Standalone Tool Examples

FSC with Gene Aggregation

krewlyzer fsc -i sample.bed.gz -o results/ \
    --assay xs2 \
    --target-regions targets.bed

Outputs: sample.FSC.tsv + sample.FSC.gene.tsv

WPS with Dual Output

krewlyzer wps -i sample.bed.gz -o results/ \
    --assay xs2 \
    --target-regions targets.bed

Outputs: sample.WPS.parquet (genome-wide) + sample.WPS.panel.parquet (panel genes)

Building Your Own PON

Create a PON from your healthy plasma samples:

# Create sample list file
ls /path/to/healthy/*-duplex.bam > healthy_samples.txt

# Build PON (HPC with custom temp directory)
krewlyzer build-pon healthy_samples.txt \
    --assay xs2 \
    --reference hg19.fa \
    --target-regions MSK-ACCESS-v2_targets.bed \
    --temp-dir /scratch/$USER/pon_tmp \
    --output xs2.pon.parquet \
    --threads 16 \
    --verbose

Tip: Use --temp-dir to specify a directory with more disk space than /tmp. Each BAM extraction creates temporary BED.gz files (~100MB each).

Nextflow Batch Processing

For multiple samples, use the Nextflow pipeline:

nextflow run nextflow/main.nf \
    --samplesheet samples.csv \
    --ref /path/to/hg19.fa \
    --outdir results/

Samplesheet Example

sample,bam,meth_bam,vcf,bed,maf,single_sample_maf,assay,pon,targets
ACCESS_001,/data/sample1.bam,,,,,false,XS2,,
ACCESS_002,/data/sample2.bam,,,,,false,XS2,,
ACCESS_003,/data/sample3.bam,,,,,false,XS2,,

When assay=XS2 is set, the pipeline automatically resolves: - PON: {asset_dir}/pon/GRCh37/xs2.pon.parquet - Targets: {asset_dir}/targets/GRCh37/xs2.targets.bed

Key Differences from WGS

Aspect WGS MSK-ACCESS
Coverage Uniform Targeted
GC Model All fragments Off-target only
FSC Output Windows + Gene-level
WPS Output Genome-wide + Panel-specific
Fragments used All Split on/off-target

Troubleshooting

"0% of reads pass filters"

Duplex/consensus BAMs need: 

krewlyzer run-all ... --no-require-proper-pair

Missing Gene BED

Ensure bundled data is present: 

ls $(python -c "from krewlyzer.assets import AssetManager; a=AssetManager('hg19'); print(a.base_path)")/genes/GRCh37/
# Should show: xs1.genes.bed.gz, xs2.genes.bed.gz

Next Steps