Skip to contents

Diagnose zero inflation

Source: R/auto_distribution.R
diagnose_zero_inflation.Rd

Tests whether a dataset has excess zeros beyond what the chosen distribution predicts, and recommends a zero-inflation mode.

Usage

diagnose_zero_inflation(data, model, threshold = 0.05)

Arguments

data

Input matrix (sparse or dense)

model

A fitted NMF model (any distribution)

threshold

Minimum excess zero fraction to declare ZI. Default 0.05.

Value

A list with:

excess_zero_rate

Fraction of zeros exceeding the expected count

has_zi

Logical: TRUE if excess_zero_rate > threshold

zi_mode

Recommended mode: "none", "row", or "col"

row_excess

Per-row excess zero rates

col_excess

Per-col excess zero rates

Details

Computes the expected number of zeros under the fitted distribution (Poisson approximation: \(P(X=0) \approx e^{-\mu}\)), compares to the observed zero count, and recommends ZI if the excess is large.

ZI granularity is determined by whether per-row and per-col excess rates have high variance (suggesting different rows/cols have different ZI levels).

See also

auto_nmf_distribution, nmf

Examples

# \donttest{
data(aml)
model <- nmf(aml, k = 3, maxit = 10, seed = 1)
zi <- diagnose_zero_inflation(aml, model)
zi$has_zi
#> [1] FALSE
# }