fit_signatures: Fit mutational signatures

In kgori/sigfit: Flexible Bayesian inference of mutational signatures

Description

fit_signatures performs MCMC sampling to fit a set of mutational signatures to a collection of mutational catalogues and estimate the exposure of each catalogue to each signature. Four models of signatures are available: multinomial, Poisson, normal and negative binomial. The normal model can be used when counts contains continuous (non-integer) values, while the negative binomial model is a more noise-robust version of the Poisson model.

Usage

fit_signatures(
  counts,
  signatures,
  exp_prior = NULL,
  model = "multinomial",
  opportunities = NULL,
  ...
)

Arguments

counts	Numeric matrix of observed mutation counts, with one row per sample and one column per mutation type.
signatures	Mutational signatures to be fitted; either a numeric matrix with one row per signature and one column per mutation type, or a list of matrices generated via retrieve_pars.
exp_prior	Numeric matrix with one row per sample and one column per signature, to be used as the Dirichlet priors for the signature exposures. Default priors are uniform.
model	Name of the model to sample from. Admits character values "multinomial" (default), "poisson", "negbin", "normal", "nmf" (an alias for "multinomial"), and "emu" (an alias for "poisson").
opportunities	Numeric matrix of optional mutational opportunities, with one row per sample and one column per mutation type. It also admits character values "human-genome" or "human-exome", in which case the mutational opportunities of the reference human genome/exome will be used for every sample.
...	Additional arguments to be passed to rstan::sampling.

Value

A list with two elements:

• `data`: list containing the input data supplied to the model.

• `result`: object of class stanfit, containing the output MCMC samples, as well as information about the model and the sampling process.

The model parameters (such as signature exposures) can be extracted from this object using retrieve_pars.

Examples

## Not run: 
# Load example mutational catalogues and COSMIC signatures
data("counts_21breast")
data("cosmic_signatures_v2")

# Fit signatures 1 to 4, using a custom prior that favors signature 1 over the rest
# (4 chains, 300 warmup iterations + 300 sampling iterations - use more in practice)
samples_1 <- fit_signatures(counts_21breast, cosmic_signatures_v2[1:4, ],
                            exp_prior = c(10, 1, 1, 1), iter = 600)

# Fit all the signatures, running a single chain for many iterations
# (3000 warmup iterations + 10000 sampling iterations)
samples_2 <- fit_signatures(counts_21breast, cosmic_signatures_v2, chains = 1,
                            iter = 13000, warmup = 3000)

## End(Not run)

kgori/sigfit documentation built on Feb. 3, 2022, 12:04 p.m.