calculate group mean score using glm and order genes based on scores difference

top_markers_glm(
  data,
  label,
  n = 10,
  family = gaussian(),
  batch = NULL,
  scale = TRUE,
  use.mgm = TRUE,
  pooled.sd = FALSE,
  softmax = TRUE,
  tau = 1
)

Arguments

data: matrix, features in row and samples in column
label: a vector of group label
n: integer, number of returned top genes for each group
family: family for glm, details in stats::glm()
batch: a vector of batch labels, default NULL
scale: logical, if to scale data by row
use.mgm: logical, if to scale data using scale_mgm()
pooled.sd: logical, if to use pooled SD for scaling
softmax: logical, if to apply softmax transformation on output
tau: numeric, hyper parameter for softmax

Value

a tibble with feature names, group labels and ordered processed scores

Details

When family is gaussian() with the identity link (the default) and the design matrix is full-rank, top_markers_glm() computes all per- gene label coefficients in a single closed-form least-squares solve via Matrix::solve(crossprod(X), crossprod(X, t(data))), avoiding the per-gene glm() loop. For any other family, or a rank-deficient design, the function automatically falls back to the legacy apply(data, 1, glm(...)) path, so results are unchanged for users who pass e.g. family = Gamma() or family = poisson().

Examples

data <- matrix(rgamma(100, 2), 10, dimnames = list(1:10))
top_markers_glm(data, label = rep(c("A", "B"), 5))
#> # A tibble: 20 × 3
#> # Groups:   .dot [2]
#>    .dot  Genes Scores
#>    <chr> <chr>  <dbl>
#>  1 A     6     0.252 
#>  2 A     5     0.137 
#>  3 A     2     0.116 
#>  4 A     10    0.111 
#>  5 A     8     0.105 
#>  6 A     1     0.0939
#>  7 A     7     0.0666
#>  8 A     3     0.0453
#>  9 A     9     0.0382
#> 10 A     4     0.0353
#> 11 B     4     0.201 
#> 12 B     9     0.186 
#> 13 B     3     0.157 
#> 14 B     7     0.107 
#> 15 B     1     0.0757
#> 16 B     8     0.0675
#> 17 B     10    0.0643
#> 18 B     2     0.0610
#> 19 B     5     0.0520
#> 20 B     6     0.0283