Manual Sigma
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 10
Package ‘SigMA’
December 5, 2018
Title Signature Multivariate Analysis
Version 1.0.0.0
Description SigMA is a signature analysis tool optimized to detect the mutational signature associ-
ated to HR defect, Signature 3, from hybrid capture panels, exomes and whole genome sequenc-
ing. For panels with low SNV counts, conventional signature analysis tools do not per-
form well while the novel approach of SigMA allows it to detect Signature 3-positive tu-
mors with 74% sensitivity at 10% false positive rate. One novelty of SigMA is a likeli-
hood based matching: We associate a new patient's mutational spectrum to subtypes of tu-
mors according to their signature composition. The subtypes of tumors are defined us-
ing the WGS data from ICGC and TCGA consortia, by a clustering of signature frac-
tions with hierarchical clustering. The likelihood of the sample to belong to each tumor sub-
type is calculated, and the likelihood of Signature 3 is the sum of the likelihoods of all Signa-
ture 3-positive tumor subtypes. The second novel step is the multivariate analysis with gradi-
ent boosting machines, which allows us to obtain a final score for presence of Signature-3 com-
bining likelihood with cosine similarity and exposure of Signature 3 obtained with non-
negativel-least-squares (NNLS) algorithm. The multivariate analysis allows us to automati-
cally handle different sequencing platforms. For different platforms different methods for signa-
ture analysis become more efficient, e.g. for WGS data it is not necessary to associate the tu-
mor to a subtype of tumors, because it is possible to determine Signature 3 with NNLS acu-
rately. We have a new feature also for these cases and we calculate the likelihood of NNLS de-
composition to be unique. This likelihood value was found to be the most influential fea-
ture in the multivariate analysis.
Depends R (>= 3.4.0)
License What license is it under?
Encoding UTF-8
LazyData true
RoxygenNote 6.1.0
Imports BSgenome,
BSgenome.Hsapiens.UCSC.hg19,
devtools,
DT,
GenomicRanges,
ggplot2,
gbm,
grid,
gridExtra,
IRanges,
nnls,
1
2assignment
reshape2,
Rmisc,
shinycssloaders,
VariantAnnotation
Rtopics documented:
assignment ......................................... 2
calc_llh ........................................... 3
cosine ............................................ 3
decompose ......................................... 4
lite_df............................................ 4
make_matrix ........................................ 5
match_to_catalog...................................... 6
plot_detailed ........................................ 6
plot_summary........................................ 7
plot_tribase_dist....................................... 7
predict_mva......................................... 8
run.............................................. 8
Index 10
assignment Assigns a boolean based on a threshold on the likelihood or mva score
for whether the signature is identified
Description
Assigns a boolean based on a threshold on the likelihood or mva score for whether the signature is
identified
Usage
assignment(df_in, method = "mva", signame = "Signature_3",
data = NULL, tumor_type = "breast", do_strict = T, weight_cf)
Arguments
df_in input data.frame
method ’median_catalog’ for likelihood based selection or ’mva’ for multivariate analy-
sis score based selection
signame name of the signature that user wants to identify, ’Signature_3’ or ’Signature_msi’
data ’msk’, ’seqcap’ or ’wgs’
tumor_type tumor type as listed in https://github.com/parklab/SigMA/ because the thresh-
olds are tumor_type specific
do_strict sets whether a strict threshold should be applied or a loose one
Value
a data.frame with a single column which contains the boolean indicating the presence of the signa-
ture
calc_llh 3
calc_llh Calculates likelihood of the genome with respect to the available sig-
nature probability distributions
Description
Calculates likelihood of the genome with respect to the available signature probability distributions
Usage
calc_llh(spectrum, signatures, counts = NULL, normalize = T)
Arguments
spectrum is the mutational spectrum
signatures is the reference signature catalog with the probability distributions
counts is the number of cases in each cluster that is represented in the catalog. They are
used as weights for each signature in the catalog
normalize is true by default only for when it is used together with NNLS in the match_to_catalog
function it is not normalized here but outside of the function
cosine calculates cosine similarity between the spectrum and a set of signa-
tures
Description
calculates cosine similarity between the spectrum and a set of signatures
Usage
cosine(x, y)
Arguments
spectrum is the mutational spectrum
signatures is the reference signature catalog with the probability distribution
4lite_df
decompose Decomposes the mutational spectrum of a genome in terms of tumor
type specific signatures that were calculated through analysis of pub-
lic WGS samples from ICGC and TCGA consortia, and contained as a
list in the package. Non-negative-least squares algorithm is used and
the number of signatures to be considered in the decomposition is in-
creased gradually, first all pairs from among the available signatures
are considered and minimal error pair is kept. Then all 3-signature
combinations, 4-signature combinations and so on are considered.
The result is updated if the error is smaller with larger number of
signatures
Description
Decomposes the mutational spectrum of a genome in terms of tumor type specific signatures that
were calculated through analysis of public WGS samples from ICGC and TCGA consortia, and
contained as a list in the package. Non-negative-least squares algorithm is used and the number of
signatures to be considered in the decomposition is increased gradually, first all pairs from among
the available signatures are considered and minimal error pair is kept. Then all 3-signature com-
binations, 4-signature combinations and so on are considered. The result is updated if the error is
smaller with larger number of signatures
Usage
decompose(spect, signatures, data)
Arguments
spect composite spectrum that is being decomposed
signatures a data.frame that contains the signatures in its columns
data sequencing platform that as in run(), used for setting the maximum number of
signatures that is allowed in the decomposition
lite_df produces a data.frame with fewer columns for easier use
Description
produces a data.frame with fewer columns for easier use
Usage
lite_df(merged_output)
Arguments
merged_output is the input data.frame
make_matrix 5
make_matrix Converts somatic mutation call files in a directory either in the form of
vcf or maf into a 96-dimensional matrix, it works for general number
of context and for 1 or 2 strands
Description
Converts somatic mutation call files in a directory either in the form of vcf or maf into a 96-
dimensional matrix, it works for general number of context and for 1 or 2 strands
Usage
make_matrix(directory, file_type = "vcf",
ref_genome = BSgenome.Hsapiens.UCSC.hg19::BSgenome.Hsapiens.UCSC.hg19,
ncontext = 3, nstrand = 1, chrom_colname = NULL,
pos_colname = NULL, ref_colname = NULL, alt_colname = NULL)
Arguments
directory pointer to the directory where input vcf maf files reside
file_type ’maf’, ’vcf’ or ’custom’
ref_genome name of the BSgenome currently set by default to BSgenome.Hsapiens.UCSC.hg19
ncontext number of bases in the nucleotide sequence which makes up the spectrum, de-
fault 3
nstrand number of strands to be considered, 1 contracts to a single strand which for
ncontext = 3 gives the commonly used 96 dimensions
chrom_colname used only for custom files a character string defining the colname which holds
the chromosome number
pos_colname used only for custom files a character string defining the colname which holds
the position information
ref_colname used only for custom files a character string defining the colname which holds
the ref allele
alt_colname used only for custom files a character string defining the colname which holds
the alt allele
Examples
by default runs on vcf input and produces 96 dimensional spectra
make_matrix(directory = 'input')
make_matrix(directory = 'input',
file_type = 'vcf',
ref_genome = BSgenome.Hsapiens.UCSC.hg19,
ncontext = 5,
nstrand = 2)
6plot_detailed
match_to_catalog Calculates the compatibility of a list of genomes to an input catalog
based on likelihood and cosine similarity
Description
Calculates the compatibility of a list of genomes to an input catalog based on likelihood and cosine
similarity
Usage
match_to_catalog(genomes, signatures, data, cluster_fractions = NULL,
method = "median_catalog")
Arguments
genomes a data table or matrix with snv spectra in the first ntype columns and genomes
in each row
signatures the input catalog, a data table with signature spectra in each column
data sets the type of sequencing platform used, options are ’msk’, ’seqcap’, ’wgs’
method can be ’median_catalog’, ’weighted_catalog’ ’cosine_simil’ or ’decompose. ’me-
dian_atalog’ uses the signature catalog formed by clustering genome SNV spec-
tra and using it as a probability distribution. The ’median_catalog’ method can
be used with any custom signatures data frame if the user intends to provide
their own signature table.
Value
A data frame that contains the input genomes and in addition columns associated to each signature
in in the catalog with likelihood and cosine simil values
plot_detailed Generates a detailed plot per sample
Description
Generates a detailed plot per sample
Usage
plot_detailed(file = NULL, sample = NULL)
Arguments
file the csv file produced by SigMA
sample name to be plotted
plot_summary 7
plot_summary Generates summary plot
Description
Generates summary plot
Usage
plot_summary(file = NULL)
Arguments
file the csv file produced by SigMA
plot_tribase_dist plots the 96 dimensional mutational spectrum
Description
plots the 96 dimensional mutational spectrum
Usage
plot_tribase_dist(df_snvs, file_name = "test.png", labely = "N SNVs",
legend = T, text_size = 10, signame = "")
Arguments
df_snvs a data frame with 96-dimensional spectra on its columns
file_name the name of the plot to be generated with the proper extension e.g. "test.pdf",
"test.png", etc
labely string for the label of the y axis
legend boolean determining whether legend should be printed
text_size size of the text of the x and y axis text and titles
signame a text to be printed on the figure
8run
predict_mva This function uses the trained MVA, in particular gradient boosting
models, inside the package to assign a probability for the existence of
the signature of interest.
Description
This function uses the trained MVA, in particular gradient boosting models, inside the package to
assign a probability for the existence of the signature of interest.
Usage
predict_mva(input, signame, data, tumor_type = "breast", weight_cf)
Arguments
input is a data frame that has likelihood cosine similarity and total snv values in it’s
columns
signame name of the signature which is being identified
data determines the sequencing platform see run()
tumor_type tumor type tag see ?run
Value
a data.frame with a single column with the score of MVA
run Runs SigMA: (1) calculates likelihood, cosine similarity, NNLS expo-
sures, and likelihood of the decomposition. (2) These features are later
used in multivariate analysis. (3) Based on scores a final decision on
existence of the signature.
Description
Runs SigMA: (1) calculates likelihood, cosine similarity, NNLS exposures, and likelihood of the
decomposition. (2) These features are later used in multivariate analysis. (3) Based on scores a final
decision on existence of the signature.
Usage
run(genome_file, output_file = NULL, do_assign = T, data = "msk",
tumor_type = "breast", do_mva = T, check_msi = F, weight_cf = F,
lite_format = F, add_sig3 = F)
run 9
Arguments
genome_file a csv file with snv spectra info can be created from vcf file using @make_genome_matrix()
function see ?make_genome_matrix
output_file the output file name, can be NULL in which case input file name is used and
appended with "_output"
do_assign boolean for whether a cutoff should be applied to determine the final decision or
just the features should be returned
data the options are "msk" (for a panel that is similar size to MSK-Impact panel with
410 genes), "seqcap" (for whole exome sequencing), "seqcap_probe" (64 Mb
SeqCap EZ Probe v3), or "wgs" (for whole genome sequencing)
tumor_type the options are "bladder", "bone_other" (Ewing’s sarcoma or Chordoma), "breast",
"crc", "eso", "gbm", "lung", "lymph", "medullo", "osteo", "ovary", "panc_ad",
"panc_en", "prost", "stomach", "thy", or "uterus". The exact correspondance of
these names can be found in https://github.com/parklab/SigMA
do_mva a boolean for whether multivariate analysis should be run
check_msi is a boolean which determines whether the user wants to identify micro-sattelite
instable tumors
weight_cf determines whether the likelihood calculation will take into account the number
of tumors in each cluster when it is F the clusters get equal weights and when
it’s T they are weighted according to the fraction of tumors in each cluster
lite_format saves the output in a lite format when set to true
add_sig3 should be set to T when the likelihood of Signature 3 is calculated for tumor
types for which Signature 3 was not discovered by NMF in their WGS data
Examples
run(genome_file = "input_genomes.csv",
data = "msk",
tumor_type = "ovary")
run(genome_file = "input_genomes.csv",
data = "seqcap",
tumor_type = "bone_other")
