andrej-fischer / EMu

An Expectation-Maximization algorithm to infer mutational signatures

How to get the EMu software

The current stable release and a pre-compiled executable of EMu for Mac OS X (64bit), can be found here.

Test EMu

To test EMu, type the following on the command line while in the main directory.

$ ./build/EMu --mut 21_breast_cancers.mutations.txt --opp 21_breast_cancers.opportunity.txt --pre ./test

Compilation

To compile EMu yourself, you need an installation of the GNU scientific library (GSL, v14 or later). On a machine without admin rights, you might need to change the paths in the Makefile to point to your local installation of GSL, if they are not in /usr/local/. The software uses openMP. If you do not wish to use openMP, remove the -fopenmp flag in the Makefile.

Finally, simply type make on the command line.

EMu usage

Command line arguments:

Required

• --mut [file] Specify a file of mutation counts (a Nsamples x Nchannels matrix)

• --opp [file/human-genome/human-exome] Specify the mutational opportunity.

  This can be either (i) the path to a flat text file of mutational opportunities (a Nsamples x Nchannels matrix) or (ii) human-genome to use the human whole genome opportunity for all samples or (iii) human-exome to use the whole human exome (female) for all samples.

• --pre [path:./out] Set the prefix used for all output files (e.g. ./here/results).

Optional

• --force [int] Force the program to use a specific number of processes.

• --mcmc [int] Run a MCMC with this number of steps to probe the posterior probability distribution for the mutational signatures and find error estimates.

• --freeze [int] Perform zero-temperature Simulated-Annealing after convergence of the EM alorithm.

• --spectra [file] Use fixed mutational spectra (a Nspectra x Nchannels matrix).

  No EM will be performed. Only the activities per sample will be inferred and the mutations assigned. Useful for localizing processes in the genome.

• --weights [file] Supply (global) process activities to be used as an informed (local) activity prior. This needs to be a (M x Nspectra) matrix, where Nsamples in --mut and --opp needs to be an integer multiple of M.

EMu output files:

• ^[pre]_[Nsp]_ml_spectra.txt The spectra found in the data using EM (Nspectra x Nchannels matrix)
• ^[pre]_[Nsp]_map_activities.txt The activities found in the data using EM (Nsamples x Nspectra matrix)
• ^[pre]_[Nsp]_assigned.txt The mutations assigned to each process (Nsamples x Nspectra matrix).
• ^[pre]_bic.txt The BIC values for the number of spectra tried.

If MCMC was called:

• ^[pre]_[Nsp]_mcmc_spectra.txt The posterior mean spectra found in the data using MCMC (Nspectra x Nchannels matrix)
• ^[pre]_[Nsp]_mcmc_activities.txt The posterior mean activities found in the data using MCMC (Nsamples x Nspectra matrix)
• ^[pre]_[Nsp]_mcmc_err.txt The posterior std.dev. for the spectra using MCMC (Nspectra x Nchannels matrix)

EMu-prepare usage

EMu-prepare is a program to create the input files for EMu.

Command line arguments for EMu-prepare:

• --mut [file] A flat text file with the mutations to be analysed.

  Each line describes one mutation (please see note below). Expected format:

  sample chromosome coordinate mutation

  sample: identifier for each sample (no white space) chomosome: integer (rename X=23,Y=24,mt=25 etc.) coordinate: one-based integer chomosome coordinate mutation: format A>T

• --chr [dir] A directory of human chromosome fasta files.

  Expected file name format: chr1.fa. Rename file names for chr X, Y, mt etc., e.g. chrX.fa -> chr23.fa. You can download the latest version of the human reference genome here.

• --cnv [file] A file with all the copy number information.

  Each line is a non-standard copy number region. Format:

    sample chromosome start stop multiplier
  

  sample: identifier for each sample (no white space) chomosome: integer (rename X=23,Y=24,mt=25 etc.) start: chromosome start coordinate of cnv region stop: chromosome stop coordinate of cnv region (if -1, then extends to the end of the chromosome) multiplier: integer (in this region, this multiplier is used to integrate the opportunity)

  Note: the default multiplier is 2. This can be changed with --default [int]. If a sample has no copy number changes, still include at least one dummy line for each sample under consideration.

• --pre [string] A path for the bin-wise output files.

  Since there will be one file for each sample and each chr, it is a good idea to send them to a separate directory.

• --bin [int] The size of the non-overlapping windows for which to get mutational/opportunity data.

• --regions [file] A file with coordinates of sequenced genomic regions.

  One region per line. Expected format: chromosome start stop

EMu-prepare output files

Assuming EMu-prepare was called with --cnv cnv.txt --mut mutations.txt:

• mutations.txt.96 The same as mutations.txt, with the mutation channel appended at the end of each line.

• mutations.txt.mut.matrix A matrix of mutation counts with no. samples rows and 96 columns. Suitable for EMu.

• mutations.txt.mut.samples The samples corresponding to each row in above file.

• cnv.txt.opp.matrix A matrix of opportunity counts with no. samples rows and 96 columns. Suitable for EMu.

• cnv.txt.opp.sample The samples corresponding to each row in above file. Check that this is the same order as in mutations.txt.mut.samples.

NOTE:

In order to translate mutations to the 96 channels, EMu-prepare reads the bases 5' and 3' to the one given in a line of mutations.txt from the hard disk. It is very useful to sort the mutations file by chromosome and coordinate (otherwise the most time will be spent moving between physical locations in the hard disk). On UNIX, this can be achieved with:

sort -k2n,2 -k3n,3 mutations.txt > mutations.sorted.txt

KNOWN BUGS

There is a known bug when openMP is compiled with the Mac OS compiler gcc version 4.2.1, which leads to random abort trap:6 crashes. If possible, compile with latest gcc version. Alternatively, you can remove the -fopenmp flag from the Makefile or set the number of threads manually to one via:

export OMP_NUM_THREADS=1; ./EMu --mut 21_breast_cancers.mutations --opp 21_breast_cancers.opportunity --pre ./target/test