ReMixT

ReMixT is a tool for joint inference of clone specific segment and breakpoint copy number in whole genome sequencing data. The input for the tool is a set of segments, a set of breakpoints predicted from the sequencing data, and normal and tumour bam files. Where multiple tumour samples are available, they can be analyzed jointly for additional benefit.

Installation

Installing from conda

The recommended method of installation for ReMixT is using conda. First install anaconda python from the continuum website. Then add my channel, and the bioconda channel, and install ReMixT as follows.

conda config --add channels https://conda.anaconda.org/dranew
conda config --add channels 'bioconda'
conda install remixt

Installing from source

Clone Source Code

To install the code, first clone from bitbucket. A recursive clone is preferred to pull in all submodules.

git clone --recursive git@bitbucket.org:dranew/remixt.git

Dependencies

To install from source you will need several dependencies. A list of dependencies can be found in the conda yaml file in the repo at conda/remixt/meta.yaml.

Build executables and install

To build executables and install the ReMixT code as a python package run the following command in the ReMixT repo:

python setup.py install

Setup ReMixT

Reference genome

Download and setup of the reference genome is automated. The default is hg19. Select a directory on your system that will contain the reference data, herein referred to as $ref_data_dir. The $ref_data_dir directory will be used in many of the subsequent scripts when running destruct.

Download the reference data and build the required indexes:

remixt create_ref_data $ref_data_dir

Mappability file

Additionally, ReMixT requires a mappability file to be generated. We have provided a workflow for generating a mappability file based on bwa alignments, for other aligners, you may want to create your own mappability workflow, see remixt/mappability/bwa/workflow.py as an example.

To create a mappability file for bwa, run:

remixt mappability_bwa $ref_data_dir

Note that this workflow will take a considerable amount of time and it is recommended you run this part of ReMixT setup on a cluster or multicore machine.

For parallelism options see the section Parallelism using pypeliner.

Running ReMixT

Input Data

ReMixT takes multiple bam files as input. Bam files should be multiple samples from the same patient, with one bam sequenced from a normal sample from that patient.

Additionally, ReMixT takes a list of predicted breakpoints detected by paired end sequencing as an additional input.

Breakpoint Prediction Input Format

The predicted breakpoints should be provided in a tab separated file with the following columns:

prediction_id
chromosome_1
strand_1
position_1
chromosome_2
strand_2
position_2

The first line should be the column names, which should be identical to the above list. Each subsequent line is a breakpoint prediction. The prediction_id should be unique to each breakpoint prediction. The chromosome_, strand_ and position_ columns give the position and orientation of each end of the breakpoint. The values for strand_ should be either + or -. A value of + means that sequence to the right of chromosome_, position_ is preserved in the tumour chromosome containing the breakpoint. Conversely, a value of - means that sequence to the left of chromosome_, position_ is preserved in the tumour chromosome containing the breakpoint.

The following table may assist in understanding the strand of a break-end. Note that an inversion event produces two breakpoints, the strand configurations for both are shown. Additionally, for inter-chromosomal events, any strand configuration is possible.

Structural Variation	Strand of Leftmost Break-End	Strand of Rightmost Break-End
Deletion	+	-
Duplication	-	+
Inversion (Breakpoint A)	+	+
Inversion (Breakpoint B)	-	-

ReMixT Command Line

Running ReMixT involves invoking a single command, remixt run. The result of ReMixT is an hdf5 file storing pandas tables.

Suppose we have the following list of inputs:

Normal sample with ID 123N and bam filename $normal_bam
Tumour sample with ID 123A and bam filename $tumour_a_bam
Tumour sample with ID 123B and bam filename $tumour_b_bam
Breakpoint table in TSV format with filename $breakpoints

Additionally, ReMixT will generate the following outputs:

Results as HDF5 file storing pandas tables with filename $results_h5
Temporary files and logs stored in directory $remixt_tmp_dir (directory created if it doesnt exist)

Given the above inputs and outputs run ReMixT as follows:

remixt run $ref_data_dir $raw_data_dir $breakpoints \
    --normal_sample_id 123N \
    --normal_bam_file $normal_bam \
    --tumour_sample_ids 123A 123B \
    --tumour_bam_files $tumour_a_bam $tumour_b_bam \
    --results_files $results_h5
    --tmpdir $remixt_tmp_dir

Note that ReMixT creates multiple jobs and many parts of ReMixT are massively parallelizable, thus it is recommended you run ReMixT on a cluster or multicore machine. For parallelism options see the section Parallelism using pypeliner.

Output File Formats

The main output file is an HDF5 store containing pandas dataframes. These can be extracted in python or viewed using the ReMixT viewer. Important tables include:

stats: statistics for each restart
solutions/solution_{idx}/cn: segment copy number table for solution idx
solutions/solution_{idx}/brk_cn: breakpoint copy number table for solution idx
solutions/solution_{idx}/h: haploid depths for solution idx

Statistics

ReMixT uses optimal restarts and model selection by BIC. The statistics table contains one row per restart, sorted by BIC. The table contains the following columns:

idx: the solution index, used to refer to solutions/solution_{idx}/* tables.
bic: the bic of this solution
log_posterior: log posterior of the HMM
log_posterior_graph: log posterior of the genome graph model
num_clones: number of clones including normal
num_segments: number of segments
h_converged: whether haploid depths estimation converged
h_em_iter: number of iterations for convergence of h
graph_opt_iter: number of iterations for convergence of genome graph copy number
decreased_log_posterior: whether the genome graph optimization stopped due to a move that decreased the log posterior

Segment Copy Number

The segment copy number table adds additional columns to the segment counts table described above, including but not limited to:

major_1
minor_1
major_2
minor_2

The columns refer to the major and minor copy number in tumour clone 1 and 2.

Breakpoints Copy Number

The breakpoint copy number table contains the following columns:

prediction_id
cn_1
cn_2

The prediction_id column matches the column of the same name in the input breakpoints file, and specifies for which breakpoint prediction the copy number is being provided. The cn_1 and cn_2 columns provide the clone specific copy number for tumour clone 1 and 2 respectively.

Haploid Depths

The haploid depths is a vector of M depths for each of the M clones including the normal. To recover cell mixture proportions, simply normalize h.

Extracting Tables as TSV files

If preferred, it is possible to extract copy number and metadata in TSV and YaML format. For results file $results_h5, extract segment copy number, breakpoint copy number and meta data to files $cn_table, $brk_cn_table, $meta_data respectively as follows:

remixt write_results \
    $results_h5 $cn_table $brk_cn_table $meta_data

ReMixT Viewer

There is an experimental viewer for ReMixT at tools/remixt_viewer_app.py. Bokeh ‘>0.10.0’ is required. To use the viewer app, organize your patient sample results files as ./patient_*/sample_*.h5. From the directory containing patient subdirectories, run the bokeh server:

bokeh-server --script $REMIXT_DIR/tools/remixt_viewer_app.py

Then navigate to http://127.0.0.1:5006/remixt.

Test Dataset for ReMixT

A test dataset is provided for providing the ability to run a quick analysis of a small dataset to ensure remixt is working correctly.

We will assume that the REMIXT_DIR environment variable points to a clone of the ReMixT source code. Additionally, create a directory, and set the environment variable WORK_DIR to the location of that directory.

First use the remixt create_ref_data sub-command to create a reference dataset. Specify a config, and use the example config that restricts to chromosome 15.

remixt create_ref_data $WORK_DIR/ref_data \
    --config $REMIXT_DIR/examples/chromosome_15_config.yaml

Use wget to retrieve a precomputed mappability file.

wget ftp://ftp.bcgsc.ca/public/shahlab/ReMixT/hg19.100.bwa.mappability.h5 --directory-prefix $WORK_DIR/ref_data/

Use wget to retrieve the example bam files and their indices for chromosome 15, and the breakpoints file with chromosome 15 breakpoints.

wget ftp://ftp.bcgsc.ca/public/shahlab/ReMixT/HCC1395*chr15.bam* --directory-prefix $WORK_DIR/
wget ftp://ftp.bcgsc.ca/public/shahlab/ReMixT/HCC1395_breakpoints.tsv --directory-prefix $WORK_DIR/

Use the remixt run sub-command to run a remixt analysis.

remixt run $WORK_DIR/ref_data $WORK_DIR/raw_data $WORK_DIR/HCC1395_breakpoints.tsv \
    --config $REMIXT_DIR/examples/chromosome_15_config.yaml \
    --tmpdir $WORK_DIR/tmp_remixt \
    --tumour_sample_ids HCC1395 \
    --tumour_bam_files $WORK_DIR/HCC1395_chr15.bam \
    --normal_sample_id HCC1395BL \
    --normal_bam_file $WORK_DIR/HCC1395BL_chr15.bam \
    --loglevel DEBUG \
    --submit local \
    --results_files $WORK_DIR/HCC1395.h5

Use the remixt write_results sub-command to write out tables of results and a yaml file containing inferred parameters and other meta data.

remixt write_results $WORK_DIR/HCC1395.h5 \
    $WORK_DIR/HCC1395_cn.tsv \
    $WORK_DIR/HCC1395_brk_cn.tsv \
    $WORK_DIR/HCC1395_info.yaml

Finally, create a visualization of the solutions using the remixt visualize_solutions sub-command.

remixt visualize_solutions $WORK_DIR/HCC1395.h5 \
    $WORK_DIR/HCC1395.html

Parallelism Using Pypeliner

ReMixT uses the pypeliner python library for parallelism. Several of the scripts described above will complete more quickly on a multi-core machine or on a cluster.

To run a script in multicore mode, using a maximum of 4 cpus, add the following command line option:

--maxjobs 4

To run a script on a cluster with qsub/qstat, add the following command line option:

--submit asyncqsub

Often a call to qsub requires specific command line parameters to request the correct queue, and importantly to request the correct amount of memory. To allow correct calls to qsub, use the --nativespec command line option, and use the placeholder {mem} which will be replaced by the amount of memory (in gigabytes) required for each job launched with qsub. For example, to use qsub, and request queue all.q and set the mem_free to the required memory, add the following command line options:

--submit asyncqsub --nativespec "-q all.q -l mem_free={mem}G"

License

ReMixT is released under the MIT License.