SeparateSignatureFromBackground: Estimate a signature from experimentally exposed spectra...

In steverozen/mSigBG: Separate Background Mutational Signature from Signature Due to a Mutagen

Estimate a signature from experimentally exposed spectra minus a background signature.

Description

We index mutation channels (e.g. ACA > AAA, ACC > AAC, ...) by j, j \in 1...96.

We index input mutational spectra by i.

Let

g = g_1, g_2,… , g_{96}, with Σ g_j = 1, be the previously determined, input background signature profile,

s^i, i \in 1, 2,… be the input spectra, from exposed samples, usually only 2 or 3,

b^i, i \in 1, 2,… be the (to-be-estimated) numbers of mutations due to the background signature in each s^i, and

t = t_1, t_2,… , t_{96}, with Σ t_j = 1, be the (to-be-estimated) target signature due to an exposure.

We want to maximize Π^iP(s^i|b^i,t)P(b^i) over b^1, b^2,… and t. (Note that the code actually minimizes the additive inverse of this.)

P(b^i) is estimated from the distribution of previously observed numbers of mutations in untreated samples, with the additional constraint that b^i ≤ |s^i|), where |s^i| is defined as the total number of mutations in spectrum s^i, i.e. |s^i| = Σ_j s^i_j, j \in 1...96.

P(s^i|b^i,t) is estimated as follows:

The expected number of mutations in each mutation category, j, is estimated as

e^i_j = g_jb^i + t_j(|s^i| - b^i).

Then P(s^i|e^i) is estimated as Π_jP(s^i_j|e^i_j).

P(s^i_j|e^i_j) is estimated from a negative binomial distribution centered on each e^i_j; see the sig.nbinom.size elements of the background.info package variables.

Usage

SeparateSignatureFromBackground(
  spectra,
  bg.sig.info,
  m.opts = NULL,
  start.b.fraction = 0.1
)

Arguments

spectra	The spectra from which to subtract the background, as a matrix or ICAMS catalog.
bg.sig.info	Information about the background signature. See background.info.
m.opts	Options to pass to nloptr.
start.b.fraction	The estimated fraction of the mutations in spectra due to the background signature.

Details

See ObjFn1.

Value

A list with the elements

inferred.target.sig

The estimated target signature as a numerical vector.

exposures.to.target.sig

The estimated total number of mutations due to the target signature in each input spectrum.

exposures.to.bg.sig

The estimated total number of mutations due to the background in each input spectrum.

message

The message element of all.opt.ret.

all.opt.ret

The entire return value from the optimization. See nloptr

steverozen/mSigBG documentation built on Nov. 4, 2022, 10:58 a.m.