intNMF: Integrative NMF clustering (intNMF)

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

Integrative NMF clustering (intNMF)

Description

Joint non-negative matrix factorisation across data sources: a single shared basis matrix W (objects \times rank) and source-specific coefficient matrices H_k minimise \sum_k \alpha_k \lVert X_k - W H_k \rVert_F^2 with all factors non-negative. Sources are weighted by \alpha_k = \max_j \overline{SS}_j / \overline{SS}_k. Object clusters are obtained by consensus of the dominant-factor assignment across restarts.

Usage

intNMF(List, k, nstart = 10, max_iter = 200, seed = NULL)

Arguments

List	A list of data matrices over the same objects, objects in rows. Matrices are shifted to be non-negative internally.
k	Number of clusters (also the factorisation rank).
nstart	Number of random restarts forming the consensus.
max_iter	Maximum multiplicative-update iterations per restart.
seed	Optional RNG seed.

Value

A list of class "intNMF" with cluster, the consensus DistM, an agnes Clust, the best-fit basis W and source weights alpha.

References

Chalise, P. & Fridley, B. L. (2017). Integrative clustering of multi-level omic data based on non-negative matrix factorization. PLoS ONE 12:e0176278. Reviewed in Zhang et al. (2022), WIREs Comput Stat 14:e1553.

See Also

M_ABC, SNF

Examples

data(mosaic_toy)
fit <- intNMF(mosaic_toy$List, k = 3, nstart = 5, seed = 1)
table(fit$cluster)

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