CoRAL - Complete Reconstruction of Amplifications with Long reads

Reference

CoRAL is a tool which utilizes aligned, single-molecule long-read data (.bam) as input, and identifies candidate ecDNA structures. A pre-print is available here: https://www.biorxiv.org/content/10.1101/2024.02.15.580594v1

Installation

CoRAL can be installed and run on most modern Unix-like operating systems (e.g. Ubuntu 18.04+, CentOS 7+, macOS).

CoRAL requires python>=3.12; we recommend using venv/conda for managing Python/pip installations.

1. Clone source

   git clone https://github.com/AmpliconSuite/CoRAL
   cd CoRAL
   

2. Install packages

   • Option 1. Install With pip.

     pip install -r requirements.txt

     Set --extra-index-url https://download.pytorch.org/whl/cpu to prevent inclusion of gigantic GPU packages.

   • Option 2. Install with poetry.

     pip install poetry
     poetry install

3. Download a Gurobi optimizer license (free for academic use)

   • Place the gurobi.lic file you download into $HOME/. This path is usually /home/username/gurobi.lic.

4. Finish installing CNVkit dependencies (recommended)

   Rscript -e 'if (!require("BiocManager", quietly = TRUE)) install.packages("BiocManager")'
   Rscript -e 'BiocManager::install("DNAcopy")'

Getting copy number calls

Before running CoRAL, you will need genome-wide copy number (CN) calls generated from your long-read data.

• If you have these already, simply ensure that they are in a .bed format like so:

  chrom start end CN

• If you don't have these then you can run CNVkit (installed as a dependency) to generate them, by doing

  ./scripts/call_cnvs.sh <input.bam> ./reference/hg38full_ref_5k.cnn <output_dir>

  This will create a file called [input].cns, which you can feed to CoRAL for it's --cn_segs argument.

Command line arguments to run CoRAL

CoRAL and its various run-modes can by used in the following manner

coral [mode] [mode arguments]

The modes are as follows:

1. seed: Identify and filter copy number gain regions where amplifications exist
2. reconstruct: Perform breakpoint graph construct and cycle decomposition on the amplified seeds.
3. plot: Create plots of decomposed cycles and/or breakpoint graph sashimi plot.
4. hsr: Identify candidate locations of chromosomal homogenously staining region (HSR) integration points for ecDNA.
5. cycle2bed: Convert the AmpliconArchitect (AA) style *_cycles.txt file to a .bed format. The AA format is also used by CoRAL.

1. seed

As the seed amplification intervals are required by the main script reconstruct mode, it is suggested the user first run seed mode to generate seed amplification intervals.

Usage: coral seed <Required arguments> <Optional arguments>

Required arguments:

• --cn-seg <FILE>, Long read segmented whole genome CN calls (.bed or CNVkit .cns file).

Optional arguments:

• --output-prefix <STRING> - Prefix of the output *_CNV_SEEDS.bed file. If not specified (by default), output the *_CNV_SEEDS.bed with the same prefix as the input *.cns file.
• --gain <FLOAT> - A minimum CN threshold (with the assumption of diploid genome) for a particular CN segment to be considered as a seed. Default is 6.0.
• --min-seed-size <INT> - Minimum size (in bp) for a CN segment to be considered as a seed. Default is 100000.
• --max-seg-gap <INT> - Maximum gap size (in bp) to merge two proximal CN segments to be considered as seed intervals. If at least two segments are merged, then they will be treated as a single candidate to be filtered with --min-seed-size, and their aggregate size will be compared with the value. Default is 300000.

2. reconstruct

Usage: reconstruct <Required arguments> <Optional arguments>

2.1 Required arguments:

• --lr-bam <FILE> - Coordinate sorted *.BAM file, with *.bai index (mapped to the provided reference genome) in the same directory.
• --cnv-seed <FILE> - *.bed file with a putative list of seed amplification intervals. The seed amplification intervals can be obtained through running seed mode, or provided manually.
• --output-dir <FOLDER> - Directory to which the output graph.txt and cycles.txt files will be written.
• --cn-seg <FILE> - Long read segmented whole genome CN calls (.bed or CNVkit .cns file).

2.2 Optional arguments:

• --min-bp-support <FLOAT> - Filter out breakpoints with less than (min_bp_support * normal coverage) long read support in breakpoint graph construction. The default value is set to 1.0, meaning to filter out breakpoints supported by less than diploid coverage, but it is highly recommended to specify a much larger value, e.g. 10.0 to obtain a cleaner breakpoint graph and the dominating ecDNA cycle(s).
• --skip-cycle-decomp - If specified, will stop by only outputting the breakpoint graph files *_graph.txt (see Expected output below) for all amplicons and not extract cycles from the graph and output *_cycles.txt.
• --output-all-path_constraints - If specified, output all path constraints given by long reads in *_cycles.txt file (see "Expected output" below).
• --cycle-decomp-alpha <FLOAT between [0, 1]> - Parameter used to balance CN weight and path constraints in the objective function of greedy cycle extraction. Default value is 0.01, higher values favor the satisfaction of more path constraints.
• --cycle-decomp-time-limit <INT> - Maximum running time (in seconds) reserved for solving the quadratic program with Gurobi (integer program solver). The solver would return the best solution(s) it currently found, regardless of the optimality status, when reaching this time limit. Default value is 7200 (i.e., 2 hours).
• --cycle-decomp-threads <INT> - Number of threads reserved for for solving the quadratic program with Gurobi (integer program solver). If not specified (and by default), the solver would attempt to use up all available cores in the working machine.
• --postprocess-greedy-sol - If specified, automatically postprocess the cycles/paths returned in greedy cycle extraction, by solving the full quadratic program to minimize the number of cycles/paths starting with the greedy cycle extraction solution (as an initial solution).
• --log-file <FILE> - Name of the main *.log file, which can be used to trace the status of reconstruct run(s).

2.3 Expected output:

CoRAL may identify and reconstruct a few distinct focal amplifications in the input *.BAM sample, each will be organized as an amplicon, which includes a connected component of amplified intervals and their connections by discordant edges. CoRAL writes the following files to the directory specified with --output_dir.

• Graph file: For each amplicon, a tab-separated text file named output_dir/amplicon*_graph.txt describing the sequence edges, concordant edges and discordant edges in the graph and their predicted copy count. Note that the graph files outputted by CoRAL have the same format as those outputted by AmpliconArchitect (and therefore the files can be used interchangeably with AmpliconArchitect). Here is an example graph file from GBM39, a cell line with EGFR amplified on ecDNA.

SequenceEdge: StartPosition, EndPosition, PredictedCN, AverageCoverage, Size, NumberOfLongReads
sequence	chr7:54659673-	chr7:54763281+	4.150534	45.907363	103609	576
sequence	chr7:54763282-	chr7:55127266+	89.340352	1052.714362	363985	40637
sequence	chr7:55127267-	chr7:55155020+	2.843655	32.729552	27754	172
sequence	chr7:55155021-	chr7:55609190+	89.340352	1013.182857	454170	49675
sequence	chr7:55609191-	chr7:55610094+	2.868261	31.027655	904	915
sequence	chr7:55610095-	chr7:56049369+	89.340352	1023.280633	439275	49106
sequence	chr7:56049370-	chr7:56149664+	4.150534	49.623899	100295	562
BreakpointEdge: StartPosition->EndPosition, PredictedCN, NumberOfLongReads
concordant	chr7:54763281+->chr7:54763282-	4.150534	26
concordant	chr7:55127266+->chr7:55127267-	2.843655	36
concordant	chr7:55155020+->chr7:55155021-	2.843655	32
concordant	chr7:55609190+->chr7:55609191-	2.697741	38
concordant	chr7:55610094+->chr7:55610095-	2.697741	41
concordant	chr7:56049369+->chr7:56049370-	4.150534	45
discordant	chr7:55610095-->chr7:55609190+	86.642611	869
discordant	chr7:56049369+->chr7:54763282-	85.189818	981
discordant	chr7:55155021-->chr7:55127266+	86.496697	978

• Cycles file: For each amplicon, a tab-separated text file named output_dir_amplicon*_cycles.txt describing the list of cycles and paths returned from cycle extraction. Note that the cycles files output by CoRAL have mostly the same format as those output by AmpliconArchitect (and therefore the files can be used interchangeably with AmpliconArchitect in most cases). Specifically a cycles file includes (i) the list of amplified intervals; (ii) the list of sequence edges; (iii) the list of cycles and paths, where an entry starts with 0+ and ends with 0- in Segments indicates a path - these lines have the same format as AmpliconArchitect output. CoRAL's cycles files additionally include (iv) a list of longest (i.e., there are no paths that can form a sub/super-path to each other) path constraint indicated by long reads, and used in CoRAL's cycle extraction. Here is an example cycles file corresponding to the above graph file from GBM39.

Interval	1	chr7	54659673	56149664
List of cycle segments
Segment	1	chr7	54659673	54763281
Segment	2	chr7	54763282	55127266
Segment	3	chr7	55127267	55155020
Segment	4	chr7	55155021	55609190
Segment	5	chr7	55609191	55610094
Segment	6	chr7	55610095	56049369
Segment	7	chr7	56049370	56149664
List of longest subpath constraints
Path constraint	1	2+,3+,4+	Support<=6	Satisfied
Path constraint	2	4+,5+,6+	Support<=34	Satisfied
Cycle=1;Copy_count=82.34616279663038;Segments=2+,4+,6+;Path_constraints_satisfied=
Cycle=2;Copy_count=2.8436550275157644;Segments=0+,2+,3+,4+,5+,6+,0-;Path_constraints_satisfied=1,2

Note that if --output-all-path-constraints is specified, then all path constraints given by long reads will be written to in *.cycles file.

• Other outputs include the output_dir_amplicon*_model.lp file(s) and output_dir_amplicon*_model.log file(s) given by Gurobi (integer program solver), for each amplicon, respectively describing the quadratic (constrainted) program in a human readable format, and the standard output produced by Gurobi.

3. plot

Usage: coral plot <Required arguments> <Optional arguments>

3.1 Required arguments: If --plot-graph is given, --graph is required. If --plot-cycles is given --cycles is required.

Argument	Description
--ref <choice>	Reference genome choice. Must be one of [hg19, hg38, GRCh38, mm10]
--bam <file>	Bam file the run was based on
--graph <file>	AA-formatted _graph.txt file
--cycles <file>	AA-formatted _cycles.txt file
--output-dir <str>	Directory for output files

3.2 Optional arguments:

Argument	Default	Description
--plot-graph		Plot the AA graph file CN, SVs and coverage as a sashimi plot
--plot-cycles		Plot the AA cycles file genome decompositions
--only-cyclic-paths		Only visualize the cyclic paths in the cycles file
--num-cycles <int>	[all]	Only plot the first [arg] cycles from the cycles file
--max-coverage <float>	[1.25x max coverage in region]	Do not extend coverage plot in graph sashimi plot above [arg] value
--min-mapq <int>	15	Do not use alignment in coverage plot with MAPQ value below [arg]
--gene-subset-list <str> <str> <str> ...	[all]	Only indicate positions of the gene names in this list
--hide-genes		Do not plot positions of genes
--gene-fontsize <float>	12	Adjust fontsize of gene names
--bushman-genes		Only plot genes found in the Bushman lab cancer-related gene list ('Bushman group allOnco').
--region <chrom:pos1-pos2>	[entire amplicon]	Only plot genome region in the interval given by chrom:start-end

4. hsr

Usage: coral hsr <Required arguments> <Optional arguments>

4.1 Required arguments:

Argument	Descripion
--lr-bam <file>	Coordinate-sorted and indexed long read .bam file
--cycles <file>	AA-formatted _cycles.txt file
--cn-segs <file>	Long read segmented whole genome CN calls (.bed or CNVkit .cns file).
--normal-cov <float>	Estimated coverage of diploid genome regions

4.2 Optional arguments:

Argument	Default	Description
--bp_match_cutoff	100	Breakpoint matching cutoff distance (bp)
--bp_match_cutoff_clustering	2000	Crude breakpoint matching cutoff distance (bp) for clustering

5. cycle2bed

CoRAL provides an option to convert its cycles output in AmpliconArchitect format *_cycles.txt into *.bed format (similar to Decoil), which makes it easier for downstream analysis of these cycles.

Usage: coral cycle2bed <Required arguments> <Optional arguments>

5.1 Required arguments:

• --cycle-file <FILE> - Input cycles file in AmpliconArchitect format.
• --output-file <FILE> - Output cycles file in *.bed format.

5.2 Optional arguments:

• --num-cycles <INT> - If specified, only convert the first NUM_CYCLES cycles.

Here is an example output of cycle2bed given by the above cycles file from GBM39.

#chr	start	end	orientation	cycle_id	iscyclic	weight
chr7	54763282	55127266	+	1	True	82.346163
chr7	55155021	55609190	+	1	True	82.346163
chr7	55610095	56049369	+	1	True	82.346163
chr7	54763282	56049369	+	2	False	2.843655