ISOpure.step1.CPE: Perform first step of ISOpure purification algorithm

Description Usage Arguments Value Author(s) References

View source: R/ISOpure.step1.CPE.R

Description

Performs the first step of the ISOpure purification algorithm, taking tumor data normal profiles and returning the a list, ISOpureS1model, with all the updated parameters.

Usage

1
ISOpure.step1.CPE(tumordata, BB, PP, MIN_KAPPA, logging.level) 

Arguments

tumordata

a GxD matrix representing gene expression profiles of heterogeneous (mixed) tumor samples, where G is the number of genes, D is the number of tumor samples.

BB

represents B = [b_1 ... b_(K-1)] matrix (from Genome Medicine paper) a Gx(K-1) matrix, where (K-1) is the number of normal profiles (β_1,...,β_(K-1)), G is the number of genes. These are the normal profiles representing normal cells that contaminate the tumor samples (i.e. normal samples from the same tissue location as the tumor). The minimum element of BB must be greater than 0 – i.e. every gene/transcript must be observed on some level in each normal sample.

PP

a GxM matrix, representing the expression profiles whose convex combination form the prior over the purified cancer profile learned.

MIN_KAPPA

(optional) The minimum value allowed for the strength parameter kappa placed over the reference cancer profile m (see Quon et al, 2013). By default, this is set to 1/min(BB), such that the log likelihood of the model is always finite. However, when the min(BB) is very small, this forces MIN_KAPPA to be very large, and can sometimes cause the reference profile m to look too much like a 'normal profile' (and therefore you may observe the tumor samples having low % cancer content estimates). If this is the case, you can try setting MIN_KAPPA=1, or some other small value. For reference, for the data presented in Quon et al., 2013, MIN_KAPPA is on the order of 10^5.

logging.level

(optional) A string that gives the logging threshold for futile.logger. The possible options are 'TRACE', 'DEBUG', 'INFO', 'WARN', 'ERROR', 'FATAL'. Currently the messages in ISOpureR are only in the categories 'INFO', 'WARN', and 'FATAL', and the default setting is 'INFO'. Setting a setting for the entire package will over-ride the setting for a particular function.

Value

ISOpureS1model, a list with the following important fields:

theta

a DxK matrix, giving the fractional composition of each tumor sample. Each row represents a tumor sample that was part of the input, and the first K-1 columns correspond to the fractional composition with respect to the Source Panel contaminants. The last column represents the fractional composition of the pure cancer cells. In other words, each row sums to 1, and element (i,j) of the matrix denotes the fraction of tumor i attributable to component j (where the last column refers to cancer cells, and the first K-1 columns refer to different 'normal cell' components). The 'cancer', or tumor purity, estimate of each tumor is simply the last column of theta.

alphapurities

tumor purities (alpha_i in paper), same as the last column of the theta variable, pulled out for user convenience.

mm

reference cancer profile, in the form of parameters of a multinomial or discrete distribution (sum of elements is 1). This is the same as the purified cancer profile that ISOLATE was designed to learn.

omega

a Mx1 vector describing the convex combination weights learned by ISOpure step 1 over the PPtranspose matrix, that when applied to the Site of Origin Panel, forms the prior over the reference cancer profile. When ISOpure step 1 is used in a similar fashion to the ISOLATE algorithm, entry i indicates the "probability" that the normal profile in the i-th column of PP is the site of origin of the secondary tumors stored in tumordata.

total_loglikelihood

log likelihood of the model

vv

(internal parameter) hyper-parameters from Dirichlet distribution, representing both mean and strength of a Dirichlet distribution over theta

kappa

(internal parameter) the strength parameter over the Dirichlet distribution over the reference cancer parameter, mm

mm_weights, theta_weights, omega_weights

(internal parameters) used in the optimization of mm, theta, and omega (instead of performing constrained optimization on these positively constrained variables directly, we optimize their logs in an unconstrained fashion.)

log_BBtranspose, PPtranspose, log_all_rates:

(internal parameters) used in the calculations of loglikelihood

MIN_KAPPA

(internal parameter) as described in the Arguments section

Author(s)

Gerald Quon, Catalina Anghel, Francis Nguyen

References

G Quon, S Haider, AG Deshwar, A Cui, PC Boutros, QD Morris. Computational purification of individual tumor gene expression profiles. Genome Medicine (2013) 5:29, http://genomemedicine.com/content/5/3/29.

G Quon, QD Morris. ISOLATE: a computational strategy for identifying the primary origin of cancers using high-thoroughput sequencing. Bioinformatics 2009, 25:2882-2889 http://bioinformatics.oxfordjournals.org/content/25/21/2882.


ISOpureR documentation built on May 2, 2018, 1:04 a.m.