Skip to contents

Calculate Adjusted Mutual Information (AMI)

Source: R/03_metrics_clustering.R
calc_ami.Rd

Computes the Adjusted Mutual Information (AMI) between two clusterings or cell-type label assignments, correcting for chance expected mutual information under the hypergeometric model with fixed marginals (Vinh et al., JMLR 2010), as used in scMoMtF (Lan et al., PLOS Comput Biol 2024).

Usage

calc_ami(
  pred,
  truth,
  average_method = c("arithmetic", "max", "min", "geometric")
)

Arguments

pred

Vector of predicted cluster labels.

truth

Vector of ground truth cell type labels.

average_method

Method to normalize mutual information: "arithmetic" (default, scikit-learn standard), "max", "min", or "geometric".

Value

AMI score in [0, 1] (adjusted for chance).

References

Vinh, N. X., Epps, J., & Bailey, J. (2010). Information theoretic measures for clusterings comparison: Variants, properties, normalization and correction for chance. Journal of Machine Learning Research, 11, 2837-2854.

Lan, W., Ling, T., Chen, Q. et al. scMoMtF: An interpretable multitask learning framework for single-cell multi-omics data analysis. PLOS Comput Biol 20(12): e1012679 (2024).

Examples

pred <- factor(rep(c("A", "B"), each = 20))
truth <- factor(rep(c("A", "B"), each = 20))
calc_ami(pred, truth)
#> [1] 1