`finaletoolkit delfi`¶

1. About¶

Calculates DELFI features over the genome, returning information about (GC-corrected) short fragments, long fragments, DELFI ratio, and total fragments.

DELFI — **DE**tection of **F**ragmentation for **L**iquid biopsy **I**nvestigation, from Cristiano et al. (2019) — measures the ratio of short to long cfDNA fragments in bins across the genome. Tumor-derived cfDNA is systematically shorter than cfDNA from healthy cells, so the spatial pattern of this ratio is one of the strongest published cfDNA cancer signals.

The measurement is GC-corrected, which is essential: bin-level GC content confounds both fragment length and coverage, so an uncorrected ratio partly reports GC bias rather than biology. DELFI is the most reference-hungry analysis in the pipeline, taking four positional files plus the gap and blacklist tracks.

2. As the pipeline runs it¶

finaletoolkit delfi bams/{sample}.sorted.bam <chrom_sizes> <ref2bit> <intervals> \
    -q <--mapscore> \
    --blacklist-file <blacklist> \
    -g <gap> \
    -o delfi/{sample}_delfi.bed \
    -w <threads> \
    -v \
    --no-merge-bins

2.1 Positional arguments¶

The four positionals must be given in this order:

input_file

The analysis BAM. value: bams/{sample}.sorted.bam

chrom_sizes

Chromosome name and length table. value: the chrom_sizes file for the selected --genome build

Bounds the region DELFI bins over.

Note

FinaleToolkit's help notes that to replicate the original Cristiano et al. methodology this file "should contain only autosomes". The bundled chrom_sizes files are full builds including sex chromosomes and chrM, so DELFI bins here span more than the original publication's autosome-only scope. Filter the file if you need strict replication.

reference_file

2bit reference sequence. value: the ref2bit file for the selected build

Used to compute per-bin GC content for the GC correction.

bins_file

Bins to compute DELFI over. value: intervals/{genome}_{size}_intervals.bed, tiled from the selected build's reference FastA at the --interval width (default 1mb)

The bins are not the publication's 100 kb merged to 5 Mb

The original DELFI methodology uses 100 kb bins merged to 5 Mb. This pipeline passes the same general-purpose interval file it gives frag-length-intervals and interval-end-motifs, so that one interval definition is shared across all three analyses. Combined with --no-merge-bins below, the output is at the --interval width throughout — 1mb by default. The values are therefore not numerically comparable to published DELFI scores without re-binning, whatever width you choose.

Because the width is now a run option rather than a fixed reference file, a run closer to the publication's binning is possible (--interval 100kb), but note that the merge step is still disabled, so the 5 Mb merge is not reproduced.

Windows with no coverage break GC correction

DELFI drops bins that overlap the build's gap annotation, then GC-corrects what remains. If no surviving bin contains any fragment — a very sparse or narrowly-targeted BAM, or one whose only reads sit inside a centromere — every bin's GC and ratio are NaN, and the correction fails with ValueError: arange: cannot compute length rather than a message naming the cause. Larger --interval widths make this less likely, since each window pools fragments from more of the genome.

2.2 Options¶

-q, --quality-threshold

Minimum mapping quality. value: --mapscore (default 20)

-b, --blacklist-file

Regions to ignore. value: the blacklist entry for the selected build, passed as --blacklist-file

ENCODE blacklist regions — anomalous mappability or signal artefacts. Excluding them prevents a handful of pathological loci from dominating the ratio. If the selected build has no blacklist entry, the flag is omitted and DELFI runs without it.

-g, --gap-file

Structural gaps to exclude. value: the gap entry for the selected build

A BED4 of centromere, telomere and short-arm annotations from the UCSC gap track. Also omitted if the build does not define it.

-M, --no-merge-bins

Keep the input bins; do not merge to 5 Mb. set by the pipeline

By default DELFI merges its bins up to --window-size (5 Mb) before reporting. The pipeline suppresses that so the output stays at the resolution of the input interval file, leaving any aggregation to downstream analysis rather than baking it in.

-w, --workers

Worker processes. value: the step's threads from config/cluster.json (default 24)

GC correction is on; the hg19 no-coverage fix is not disabled

The pipeline passes neither -G, --no-gc-correct (so GC correction is applied) nor -R, --keep-nocov. -R exists to skip the removal of two hg19-specific no-coverage regions and FinaleToolkit's help advises setting it for non-hg19 references; the pipeline leaves it unset for both builds, so those two regions are removed regardless of build. The affected span is negligible relative to a genome-wide bin set.

3. Output¶

delfi/{sample}_delfi.bed — one row per bin with its short-fragment count, long-fragment count, DELFI ratio and total fragment count, GC-corrected.

4. Controlling it¶

What	How
Bin resolution	`--interval`, which sets the width the bins are tiled at
Blacklist / gap exclusion	the `blacklist` and `gap` entries — remove a key to drop the flag
Threads / memory / walltime	the `delfi` entry of `config/cluster.json`

"delfi": {
    "threads": 24,
    "mem": "16G",
    "time": "08:00:00",
    "partition": "norm"
}

Note that DELFI does not take --fragment-minimum/--fragment-maximum — it defines short and long internally, since that partition is the measurement.

5. Requires¶

ref2bit, reference_fa (which the bins are tiled from) and chrom_sizes for the selected genome build. blacklist and gap are optional; each is passed only if present. Both bundled builds provide all five.

6. Reference¶

Cristiano S, et al. Genome-wide cell-free DNA fragmentation in patients with cancer. Nature, 2019;570:385–389.

finaletoolkit delfi¶