Seeking Clones {CloneSeeker}

Seeking Tumor Clones From Data

Description

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.

Usage

seekClones(cndata, vardata, cnmodels, psiset, pars, imputedCN = NULL)
runAlg(...)

Arguments

cndata

A data frame with seven columns; can also be NULL. The names of the required columns are enumerated in the man page for generateTumorData. These are the same as the output typically produced by the DNAcopy algorithm, where each row represents a segment (a contiguous region of a chromosome) where the copy number is believed to be constant.

vardata

A data frame with eight columns; can also be NULL. The names of the required columns are enumerated in the man page for generateTumorData. These are typical of the output of a DNA sequencing experiment that has been processed to identify variants, which may be either germline or somatic.

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

Details

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:

sigma0

The standard deviation of measured allelic copy number at the SNP level.

ktheta

The probability parameter of the geometric prior distribution on K, the number of clones.

theta

The probability parameter of the geometric prior distribution on genomic copy number.

mtheta

The probability parameter of the geometric prior distribution on the occurence of point mutations.

alpha

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.

thresh

The threshold determining the smallest possible detectable clone.

cutoff

SNP array segments with fewer markers than this are excluded.

Q

Determines the number of new psi vectors resampled from the estimated posterior probability distribution at each iteration of the algorithm

iters

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.

Value

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

Author(s)

Kevin R. Coombes [e-mail], Mark Zucker [e-mail]

References

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.

Examples

# 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)