classify_spectrum: Simple spectrum classifier based on empirical thresholds

In kkrismer/transite: RNA-binding protein motif analysis

Simple spectrum classifier based on empirical thresholds

Description

Spectra can be classified based on the aggregate spectrum classifier score. If sum(score) == 3 spectrum considered non-random, random otherwise.

Usage

classify_spectrum(
  adj_r_squared,
  degree,
  slope,
  consistency_score_n,
  n_significant,
  n_bins
)

Arguments

adj_r_squared	adjusted R^2 of polynomial model, returned by score_spectrum
degree	degree of polynomial, returned by score_spectrum
slope	coefficient of the linear term of the polynomial model (spectrum "direction"), returned by score_spectrum
consistency_score_n	number of performed permutations before early stopping, returned by score_spectrum
n_significant	number of bins with statistically significant enrichment
n_bins	number of bins

Value

a three-dimensional binary vector with the following components:

coordinate 1	adj_r_squared >= 0.4
coordinate 2	consistency_score_n > 1000000
coordinate 3	n_significant >= floor(n_bins / 10)

See Also

Other SPMA functions: run_kmer_spma(), run_matrix_spma(), score_spectrum(), subdivide_data()

Examples

n_bins <- 40

# random spectrum
random_sp <- score_spectrum(runif(n = n_bins, min = -1, max = 1),
  max_model_degree = 1)
score <- classify_spectrum(
  get_adj_r_squared(random_sp), get_model_degree(random_sp),
  get_model_slope(random_sp), get_consistency_score_n(random_sp), 0, n_bins
)
sum(score)

# non-random linear spectrum with strong noise component
signal <- seq(-1, 0.99, 2 / 40)
noise <- rnorm(n = 40, mean = 0, sd = 0.5)
linear_sp <- score_spectrum(signal + noise, max_model_degree = 1,
  max_cs_permutations = 100000)
score <- classify_spectrum(
  get_adj_r_squared(linear_sp), get_model_degree(linear_sp),
  get_model_slope(linear_sp), get_consistency_score_n(linear_sp), 10, n_bins
)
sum(score)
## Not run: 
# non-random linear spectrum with weak noise component
signal <- seq(-1, 0.99, 2 / 40)
noise <- rnorm(n = 40, mean = 0, sd = 0.2)
linear_sp <- score_spectrum(signal + noise, max_model_degree = 1,
  max_cs_permutations = 100000)
score <- classify_spectrum(
  get_adj_r_squared(linear_sp), get_model_degree(linear_sp),
  get_model_slope(linear_sp), get_consistency_score_n(linear_sp), 10, n_bins
)
sum(score)

## End(Not run)

# non-random quadratic spectrum with strong noise component
signal <- seq(-1, 0.99, 2 / 40)^2 - 0.5
noise <- rnorm(n = 40, mean = 0, sd = 0.2)
quadratic_sp <- score_spectrum(signal + noise, max_model_degree = 2,
  max_cs_permutations = 100000)
score <- classify_spectrum(
  get_adj_r_squared(quadratic_sp), get_model_degree(quadratic_sp),
  get_model_slope(quadratic_sp),
  get_consistency_score_n(quadratic_sp), 10, n_bins
)
sum(score)
## Not run: 
# non-random quadratic spectrum with weak noise component
signal <- seq(-1, 0.99, 2 / 40)^2 - 0.5
noise <- rnorm(n = 40, mean = 0, sd = 0.1)
quadratic_sp <- score_spectrum(signal + noise, max_model_degree = 2)
score <- classify_spectrum(
  get_adj_r_squared(quadratic_sp), get_model_degree(quadratic_sp),
  get_model_slope(quadratic_sp),
  get_consistency_score_n(quadratic_sp), 10, n_bins
)
sum(score)

## End(Not run)

kkrismer/transite documentation built on Feb. 9, 2024, 3:23 a.m.