M_ABCdeep: Multi-source M-ABC with a deep-learning (autoencoder) Step 4...

In MosaiClusteR: An Umbrella Framework for Multi-Source and Multi-Omics Clustering

Multi-source M-ABC with a deep-learning (autoencoder) Step 4 (M-ABCdeep)

Description

The deep-learning counterpart of M_ABC: runs ABCdeep.SingleInMultiple per source (autoencoder embedding + latent clustering for the base partitions), row-stacks the per-iteration label matrices, and builds one consensus dissimilarity via f.clustABC.MultiSource. The ABC algorithm of \insertCiteAmaratunga2008MosaiClusteR is preserved end to end; only Step 4 is deep. There is no Python/torch dependency and no PCA fallback (see ABCdeep.SingleInMultiple).

Usage

M_ABCdeep(
  List,
  weighting = rep("var", length(List)),
  normalize = rep(FALSE, length(List)),
  gr = c(),
  bag = rep(TRUE, length(List)),
  numsim = 500,
  numvar = rep(100, length(List)),
  NC = NULL,
  dissimilarity = "mabc",
  final_linkage = "ward",
  alpha = 0.625,
  mds = FALSE,
  topk_frac = 0.2,
  latent_dim = "auto",
  hidden_units = 32L,
  ae_epochs = 80L,
  ae_lr = 0.001,
  cluster_in_latent = "ward",
  linkage = rep("ward.D2", length(List)),
  nugget_type = "between",
  nugget_args = list(),
  min_samples = 40L,
  seed = NULL
)

Arguments

List	A list of K data matrices over the same objects, objects (samples) in rows and features in columns.
weighting, normalize, bag, numvar	Per-source vectors (recycled to K) forwarded to ABCdeep.SingleInMultiple.
gr	Unused; kept for compatibility.
numsim	Iterations per source.
NC	Base-cluster count per source (NULL = data-driven).
dissimilarity	Consensus dissimilarity, "mabc" or "abc".
final_linkage, alpha	Final clustering controls.
mds	Logical; MDS of the consensus dissimilarities.
topk_frac	Fraction of top-weighted features the autoencoder focuses on (default 0.2); forwarded to ABCdeep.SingleInMultiple.
latent_dim, hidden_units, ae_epochs, ae_lr, cluster_in_latent, linkage	Autoencoder / latent-clustering controls (scalar or length-K), forwarded to ABCdeep.SingleInMultiple.
nugget_type, nugget_args	Forwarded to the nugget weighting.
min_samples	Warn (per source) when N is below this; the autoencoder is underdetermined at very small N (default 40).
seed	Optional RNG seed.

Value

A list of class "M_ABC" with DistM and Clust.

References

\insertAllCited

See Also

ABCdeep.SingleInMultiple, M_ABC, f.clustABC.MultiSource

Examples

data(mosaic_toy)
fit <- M_ABCdeep(mosaic_toy$List, numsim = 40, NC = 3, ae_epochs = 40, seed = 1)
table(stats::cutree(stats::as.hclust(fit$Clust), 3))

MosaiClusteR documentation built on Oct. 25, 2026, 1:08 a.m.