| Title: | Seeking and Finding Clones in Copy Number and Sequencing Data |
|---|---|
| Description: | Defines the classes and functions used to simulate and to analyze data sets describing copy number variants and, optionally, sequencing mutations in order to detect clonal subsets. See Zucker et al. (2019) <doi:10.1093/bioinformatics/btz057>. |
| Authors: | Kevin R. Coombes, Mark Zucker |
| Maintainer: | Kevin R. Coombes <[email protected]> |
| License: | Apache License (== 2.0) |
| Version: | 1.0.11 |
| Built: | 2024-11-19 06:40:21 UTC |
| Source: | CRAN |
Starting with copy number segmentation data and/or sequencing mutation data for a tumor, seek the number of clones, the fraction of cells belonging to each clone, and the likely set of abnormalities in each clone.
seekClones(cndata, vardata, cnmodels, psiset, pars, imputedCN = NULL) runAlg(...)seekClones(cndata, vardata, cnmodels, psiset, pars, imputedCN = NULL) runAlg(...)
cndata |
A data frame with seven columns; can also be NULL. The
names of the required columns are enumerated in the man page for
|
vardata |
A data frame with eight columns; can also be NULL. The
names of the required columns are enumerated in the man page for
|
cnmodels |
A matrix. Each row represents a model to be considered; each column represents a clone. The entries are integers specifying the number of DNA copies present in that clone. See details. |
psiset |
A matrix. Each column represents a clone, and each row represents a different possible model of the fraction of cells per clone. See details. |
pars |
A list of algorithm parameters; see details. |
imputedCN |
a logical value; if missing, should the copy number be imputed from the mutation data. |
... |
additional variables |
The algorithm starts with an initial set of 'psi' parameters (representing the fraction of tumor belonging to each clone). It computes the best (maximum a posteriori) clonal copy number and/or number of mutated alleles for each clone for each segment/mutation, conditional on the data and each of the initial psi vectors. It then computes the posterior probability for each psi-vector and its computed copy number and mutation parameters. It uses these posterior probabilities to resample new possible psi-vectors. The process repeats iteratively, and with each iterations obtains a better estimate of psi and the clonal segment copy number and mutation assignments until it terminates.
The set of copy number models that we use is typically
generated using the following command:
as.matrix(expand.grid(lapply(1:5, function(i){0:5})))
This setup considers all (7776) possible models with up to five clones,
where the copy number for each clone ranges from 0 to 5. (In the future,
we are likely to make this the default; right now, you have to generate
these models yourself.)
The set of possible psi-vectors (that is, the fraction of cells allocated
to each clone) that we use is typically generated using the following
command:
psis.20 <- generateSimplex(20,5)
This setup considers all (192) possible divisions of the tumor into up to
five clones, where the fraction of cells per clone is any possible
multiple of 0.05. Each row is sorted to put the most abundant clones first,
which makes it easier to identify specific clones, except in the rare case
when two clones contain exactly the same fraction of cells. (In the future,
we are likely to make this the default; right now, you have to generate
these models yourself.)
The object pars is a list of numerical algorithm parameters. The
elements are:
The standard deviation of measured allelic copy number at the SNP level.
The probability parameter of the geometric prior
distribution on K, the number of clones.
The probability parameter of the geometric prior distribution on genomic copy number.
The probability parameter of the geometric prior distribution on the occurence of point mutations.
The (repeated) alpha parameter of a symmetric Dirichlet distributed prior on the fractions of cells belong to each clone; default value is 0.5, giving a Jeffreys Prior.
The threshold determining the smallest possible detectable clone.
SNP array segments with fewer markers than this are excluded.
Determines the number of new psi vectors resampled
from the estimated posterior probability distribution at each
iteration of the algorithm
The number of iterations in the algorithm.
The default settings we used are from commonly used unfinformative priors (e.g., alpha=0.5 for the Dirichlet distribution is the Jeffreys Prior) or based on empirical assessments of the variation in data (sigma0, for example, which describes variation in SNP array data).
Note that runAlg (an alias for seekClones) is DEPRECATED.
The seekClones function returns a (rather long) list containing:
psi |
The most likely posterior psi-vector, given the data. The number of non-zero entries is the number of clones found, and the non-zero entries are the fraction of cells per clone |
A |
The most likely copy numbers for the A allele in each segment in each clone. |
B |
The most likely copy numbers for the B allele in each segment in each clone. |
psibank |
A matrix, where each row is one of the psi-vectors considered during the analysis. |
psiPosts |
A numeric vector, the (marginal) posterior probability of each psi-vector considered during the analysis. |
indices |
??? |
data |
a list with two data-frame components containing the data used during the analysis. |
filtered.data |
a list with two data-frame components containing the filtered data used during the analysis. Filtering removes non-informative segments that have normal copy number or contain only germline mutations. |
etaA |
A vector of the weighted average allelic copy number for the 'A-Allele' at each segment (that is, the sum of the clonal A-allelic copy number values multiplied by the fraction of the tumor made up by each clone) |
etaB |
A vector of the weighted average allelic copy number for the 'B-Allele' at each segment |
etaM |
A vector of the weighted average number of copies of the mutated allele at each mutation |
mutated |
A matrix of the number of mutated alleles at each locus in each clone, where the number of rows is the number of somatic mutations in the data and the number of columns is the number of clones |
Kevin R. Coombes [email protected], Mark Zucker [email protected]
Zucker MR, Abruzzo LV, Herling CD, Barron LL, Keating MJ, Abrams ZB, Heerema N, Coombes KR. Inferring Clonal Heterogeneity in Cancer using SNP Arrays and Whole Genome Sequencing. Bioinformatics. To appear. doi: 10.1093/bioinformatics/btz057.
# set up models psis.20 <- generateSimplex(20,5) cnmodels <- as.matrix(expand.grid(lapply(1:5, function(i){ 0:5 }))) # set up algortihm parameters pars <- list(sigma0=5, theta = 0.9, ktheta = 0.3, mtheta = 0.9, alpha = 0.5, thresh = 0.04, cutoff = 100, Q = 100, iters = 4) # create a tumor psis <- c(0.6, 0.3, 0.1) # three clones tumor <- Tumor(psis, rounds = 100, nu = 0, pcnv = 1, norm.contam = FALSE) # simulate a dataset dataset <- generateTumorData(tumor, 10000, 600000, 70, 25, 0.15, 0.03, 0.1) result <- seekClones(dataset$cn.data, dataset$seq.data, cnmodels, psis.20, pars = pars, imputedCN = NULL)# set up models psis.20 <- generateSimplex(20,5) cnmodels <- as.matrix(expand.grid(lapply(1:5, function(i){ 0:5 }))) # set up algortihm parameters pars <- list(sigma0=5, theta = 0.9, ktheta = 0.3, mtheta = 0.9, alpha = 0.5, thresh = 0.04, cutoff = 100, Q = 100, iters = 4) # create a tumor psis <- c(0.6, 0.3, 0.1) # three clones tumor <- Tumor(psis, rounds = 100, nu = 0, pcnv = 1, norm.contam = FALSE) # simulate a dataset dataset <- generateTumorData(tumor, 10000, 600000, 70, 25, 0.15, 0.03, 0.1) result <- seekClones(dataset$cn.data, dataset$seq.data, cnmodels, psis.20, pars = pars, imputedCN = NULL)
Utility functions for working with vectors of clonal fractions.
sampleSimplex(n, d = 5) generateSimplex(k, d, reps = 1)sampleSimplex(n, d = 5) generateSimplex(k, d, reps = 1)
d |
an integer, the dimension of the simplex, or the number of clones. |
n |
an integer, the number of vectors to sample randomly. |
k |
an integer, the number of equally spaced points to select along each side of the simplex while constructing a lattice. |
reps |
an integer, the number of times to repeat the lattice. |
When studying the clonal subpopulations of a tumor sample, we
frequently need access to vectors that contain the fraction of cells
belonging to each clone. These vectors are characterized by the fact
that each entry is nonzero and they must add up to 1. The set of such
vectors/points in d-dimensional space defines the "d-simplex". The
functions defined here allow us to work with d-simplices, either by
randomly sampling vectors (sampleSimplex) or by systematically
filling the space with a regular lattice (generateSimplex).
Both functions return a matrix with d columns. Each row
contains nonzeo real numbers that sum to 1. The generateSimplex
function ensures that (a) each row is unique and (b) the entries in
each row appear in decreasing order.
Kevin R. Coombes [email protected], Mark Zucker [email protected]
sampleSimplex(5, 3) generateSimplex(5, 3)sampleSimplex(5, 3) generateSimplex(5, 3)
Simulating copy number segmentation data and sequencing mutation data for tumors composed of multiple clones.
generateTumorData(tumor, snps.seq, snps.cgh, mu, sigma.reads, sigma0.lrr, sigma0.baf, density.sigma) plotTumorData(tumor, data) tumorGen(...) dataGen(tumor, ...)generateTumorData(tumor, snps.seq, snps.cgh, mu, sigma.reads, sigma0.lrr, sigma0.baf, density.sigma) plotTumorData(tumor, data) tumorGen(...) dataGen(tumor, ...)
tumor |
an object of the |
snps.seq |
an integer; the total number of germline variants and somatic mutations to simulate in the tumor genome. |
snps.cgh |
an integer; the number of single nucleotide polymorphisms (SNPs) to simulate as meaurements made to estimate copy number. |
mu |
an integer; the average read depth of a simulated sequencing study giving rise to mutations. |
sigma.reads |
a real number; the standard deviation of the number of simulated sequencing reads per base. |
sigma0.lrr |
a real number; the standard deviation of the simulated per-SNP log R ratio (LRR) for assessing copy number. |
sigma0.baf |
a real number; the standard deviation of the simulated B allele frequency (BAF) for assessing copy number. |
density.sigma |
a real number; the standard deviation of a beta distribution used to simulate the number of SNP markers per copy number segment. |
data |
a list containing two data frames, |
... |
additional variables |
Copy number and mutation data are simulated essentially independently. Each simulation starts with a single "normal" genome, and CNVs and/or mutations are randomly generated for each new "branch" or subclone. (The number of subclones depends on the input parameters.) Each successive branch is randomly determined to descend from one of the existing clones, and therefore contains both the aberrations belonging to its parent clone and the novel aberrations assigned to it. Depending on input parameters, the algorithm can also randomly select some clones for extinction in the process of generating the heterogeneous tumor, to yield a more realistic population structure.
Note that tumorGen (an alias for Tumor that returns a
list instead of a Tumor object) and dataGen (an alias for
generateTumorData) are DEPRECATED.
The generateTumorData function returns a list with two
components, cn.data and seq.data. Each component is
itself a data frame. Note that in some cases, one of these data frames
may have zero rows or may be returned as an NA.
The cn.data component contains seven columns:
chrthe chromosome number;
seqa unique segment identifier;
LRRsimulated segment-wise log ratios;
BAFsimulated segment-wise B allele frequencies;
X and Y
simulated intensities for two separate alleles/haplotypes per segment; and
markersthe simulated number of SNPS per segment.
The seq.data component contains eight columns:
chrthe chromosome number;
seqa unique "segment" identifier;
mut.ida unique mutation identifier;
refCounts and varCounts
the simulated numbers of reference and variant counts per mutation;
VAFthe simulated variant allele frequency;
totalCountsthe simulated total number of read counts; and
statusa character (that should probably be a factor) indicating whether a variant should be viewed as somatic or germline.
The plotTumorData function invisibly returns its data
argument.
Kevin R. Coombes [email protected], Mark Zucker [email protected]
psis <- c(0.6, 0.3, 0.1) # three clones # create tumor with copy number but no mutation data tumor <- Tumor(psis, rounds = 400, nu = 0, pcnv = 1, norm.contam = FALSE) # simulate the dataset dataset <- generateTumorData(tumor, 10000, 600000, 70, 25, 0.15, 0.03, 0.1) #plot it plotTumorData(tumor, dataset)psis <- c(0.6, 0.3, 0.1) # three clones # create tumor with copy number but no mutation data tumor <- Tumor(psis, rounds = 400, nu = 0, pcnv = 1, norm.contam = FALSE) # simulate the dataset dataset <- generateTumorData(tumor, 10000, 600000, 70, 25, 0.15, 0.03, 0.1) #plot it plotTumorData(tumor, dataset)
A class that represents tumors, thought of as a collection of (sub)clones each with an associated measure as a fraction of all tumor cells.
Tumor(psi, rounds, nu = 100, pcnv = 0.5, norm.contam = FALSE, cnmax = 4) getClone(tumor, i)Tumor(psi, rounds, nu = 100, pcnv = 0.5, norm.contam = FALSE, cnmax = 4) getClone(tumor, i)
psi |
a numeric vector containing non-negative values. |
rounds |
an integer; the number of generations through which to evolve the potential clones. |
nu |
an integer; the expected number of mutations in each clonal generation. |
pcnv |
a real number between 0 and 1; the probability of a CNV occuring per generation. |
norm.contam |
a logical value; should we treat one of the cell populations as normal cells that are "contaminating" the tumor specimen? |
cnmax |
an integer, the maximum copy number allowed in the simulated data. |
tumor |
an object of the |
i |
a integer; which clone to extract. |
The Tumor class is used to represent compex tumors, each of
which consists a set of subclones representing different fractional
parts of the tumor. Each clone is characterized by a set of copy
number variants (modeled by the output produced by something like the
DNAcopy package) and, optionally, a set of sequence
mutations. Each of these genetic events is mapped to a specific
interval or point in the human genome.
In the current implementation, a Tumor consists of a weight
vector that specifies the fractions of cells for each clone and a list
of clones. At present, each clone is itself a list containing one (if
there are no mutations) or two (if there are both copy number
variants and mutations) data frames. This structure is likely to
change in later versions of the package, since we expect to implement a
full-fledged S4 class to represent clones. So, one should not rely on
the current implementation.
The constructor returns a valid object of the Tumor
class.
Although objects can be created using new, the preferred method
is to use the constructor function, Tumor.
psi a WeightVector containing
non-negative values whose sum equals one.
clonesa list, each of whose elements represents a clone.
Convert the
Tumor object into a simple numeric vector. Never
actually used in this form, since the preferred method is to write
as(WV, "list").
Kevin R. Coombes [email protected], Mark Zucker [email protected]
showClass("Tumor") tumor <- Tumor(c(0.5, 0.3, 0.2), 100) simpleTumor <- as(tumor, "list") reformed <- as(simpleTumor, "Tumor")showClass("Tumor") tumor <- Tumor(c(0.5, 0.3, 0.2), 100) simpleTumor <- as(tumor, "list") reformed <- as(simpleTumor, "Tumor")
A class, with validity checking, to contain vectors of non-negative real numbers whose sum equals one.
WeightVector(phi)WeightVector(phi)
phi |
a numeric vector containing non-negative values. |
When trying to simulate or model biological data from (sub)clonal
populations of cells, we need vectors that keep track of the fraction
of cells belonging to each clone. These vectors can only contain
non-negative entries, and the entries must add up to one. (We
thought about calling these things "ClonalFractions", but that seems
overly specialized for a notion that is likely to prove useful in
other contexts.) Such vectors of length d can also be viewed as
points of a d-dimensional simplex.
We have implemented WeightVectors as an S4 class, primarily so
we can enforce the defining properties. We also expect this design to
make it easier to use them as slots in other classes.
The constructor returns a valid object of the WeightVector
class.
Although objects can be created using new, the preferred method
is to use the constructor function, WeightVector.
psia vector containing non-negative values whose sum equals one.
Convert the
WeightVector object into a simple numeric vector. Never
actually used in this form, since the preferred method is to write
as(WV, "numeric").
Kevin R. Coombes [email protected], Mark Zucker [email protected]
showClass("WeightVector") WeightVector(1:4) as(WeightVector(c(2, 3, 5)), "numeric")showClass("WeightVector") WeightVector(1:4) as(WeightVector(c(2, 3, 5)), "numeric")