Count bam lines determines the read depth ratios of the supplied tumor and reference genomes.

The main steps are:

• reads are apportioned to windows to give raw depths and GC proportions per window
• filtering is applied to exclude windows that have unreliable data
• raw results are normalised to account for GC bias and other confounders
• a piecewise constant fitting algorithm is applied to segment the chromosomes to regions of constant copy number.

Each read is assigned proportionally to the windows that it overlaps. Reads of quality less than 10 (configurable) are ignored, as are unmapped, duplicated, secondary and supplementary reads. The raw count of bases from reads overlapping each window is divided by the window length (1000) to give a raw depth.

The proportion of Gs and Cs in bases of reads within a window is determined. For low-depth windows (raw depth < 1.0), however, this value is overridden with the GC value from the supplied GC profile data.

Certain windows overlap pseudo genes on chromosomes 15 and 16 that produce artefactually high copy numbers. These windows are excluded.

In tumor-only whole-genome mode, diploid regions can be specified, and windows that are outside of these regions are excluded.

In target region mode, only those windows that fall within the specified target regions are retained.

Windows with GC proportions outside the range [0.24,0.67] (configurable) are excluded.

Windows with a mappability value less than 0.85 (configurable) are excluded.

Each window is assigned to a bucket according to its GC proportion. Median read depths for allosome reads are calculated for each bucket. The median read depths for the buckets are smoothed by averaging the value for each bucket with those of its two neighbors. This produces a read depth normalisation factor for each bucket. The read depth of each window is normalised by the factor for its associated GC bucket. This produces a 'ratio' value which is the subject of all later normalisation steps and which is the value used to detect copy number changes.

In targeted mode, the ratio for each window is divided by a supplied enrichment factor.

The median and mean read depths across all windows are calculated. The quotient of these values is then applied as a normalisation factor across all windows.

In whole-genome mode, windows are normalised to remove mega-base scale GC biases. This normalization assumes that the median ratio of each 10Mb window (minimum 1Mb readable) should be diploid for autosomes and haploid for male sex chromosomes, in addition to the following exceptions:

The final normalisation step is to normalise so that in the output file, the mean of the ratios for the windows is 1.0.

Sparse information in COBALT may cause a noisy fit for low pass WGS. Therefore, we consolidate buckets to try to reach a median read depth of at least 8 per bucket. The ConsolidatedBucketSize is set to clamp(roundToOneSigDigit(80 / medianTumorReadCount, 10, 1000). This formula allows consolidation into buckets of up to 1000 depth windows. For standard WGS this should have no effect as medianTumorReadDepth >> 8. We should never consolidate across regions of more than 3Mb (so never across centromere). Consolidation is not used in targeted sequencing mode.

For the consolidated buckets, the mean GC ratio for both tumor and reference is calculated for the consolidated bucket and set to the centre window in the consolidated bucket.
The other buckets are masked.

Finally, a piecewise constant fit (segmentation) of the ratios is found using a penalized least squares regression algorithm ported from the Bioconductor copynumber package.

If preferred, the original copynumber program can be used for segmentation. To do this, apply the use_old_segmenter parameter. This will require prior installation of R or RStudio and then installation of the required R packages with the following R commands:

    library(BiocManager)
    install("copynumber")
    install("dplyr")

All resource files for this tool and the WiGiTs pipeline are available for download via the HMF Resource page.

In the absence of a reference bam and reference COBALT will be run in tumor-only mode.

Without a means to determine which regions of the normal are diploid, a bed file specifying these locations must be included with the tumor-only-diploid-bed parameter. A 37 bed file (DiploidRegions.37.bed.gz) and 38 equivalent are available to download from HMF-Pipeline-Resources. To create this bed file we examined the COBALT output of 100 samples.
We considered each 1000 base region to be diploid if 50% or more of the samples were diploid (0.85 >= referenceGCDiploidRatio <= 1.15 ) at this point.

As no reference data is supplied, COBALT does not try to determine gender or any chromosomal aberrations. No reference pcf file will be created. The output reference ratios will be 1 or -1 on all chromosomes even if they are allosomes. Downstream, PURPLE will adjust the allosome ratios according to the AMBER gender.

In the absence of a tumor bam and tumor COBALT will be run in germline mode. Counts and ratios are only calculated and fitted for the reference sample.

COBALT may be run on targeted data. For more information on how to run hmftools in targeted mode please see here

COBALT requires Java 17+ and for tumor-only mode can be run with the minimum set of arguments as follows:

java -jar -Xmx8G cobalt.jar \
    -tumor SAMPLE_ID \
    -tumor_bam /sample_data/SAMPLE_ID.bam \ 
    -output_dir /sample_data/ \ 
    -threads 10 \ 
    -ref_genome_version 37
    -gc_profile /ref_data/GC_profile.1000bp.37.cnp

For in tumor-reference mode the reference sample must be supplied:

java -jar -Xmx8G cobalt.jar \
    -reference SAMPLE_ID_R \
    -reference_bam /sample_data/SAMPLE_ID_R.bam \ 
    -tumor SAMPLE_ID \
    -tumor_bam /sample_data/SAMPLE_ID.bam \ 
    -output_dir /sample_data/ \ 
    -threads 10 \ 
    -ref_genome_version 37
    -gc_profile /ref_data/GC_profile.1000bp.37.cnp

A compressed copy of the GC Profile file used by HMF (GC_profile.1000bp.37.cnp) is available to download from HMF-Pipeline-Resources. This file contains 5 columns for each 1kb window of the genome {chromosome,position,GC Proportion,Non N Proportion,Average Mappability}.
A 38 equivalent is also available. Please note the downloaded file must be un-compressed before use.

COBALT supports both BAM and CRAM file formats. If using CRAM, the ref_genome argument must be included.

Performance numbers were taken from a 72 core machine using COLO829 data with an average read depth of 35 and 93 in the normal and tumor respectively. Elapsed time is measured in minutes. CPU time is minutes spent in user mode. Peak memory is measure in gigabytes.

The following tab delimited files are written:

/run_dir/cobalt/TUMOR.cobalt.ratio.tsv.gz

/run_dir/cobalt/TUMOR.cobalt.ratio.pcf

/run_dir/cobalt/REFERENCE.cobalt.ratio.pcf

TUMOR.cobalt.ratio.tsv.gz contains the counts and ratios of the reference and tumor and the GC proportions:

TUMOR.cobalt.ratio.pcf and REFERENCE.cobalt.ratio.pcf contain the segmented regions determined from the ratios.