Clustering individuals using INMTD: a novel versatile multi-view embedding framework integrating omics and imaging...

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Multi-view clustering integrates diverse data types for comprehensive individual analysis.
• Nonnegative matrix tri-factorization (NMTF) and nonnegative Tucker decomposition (NTD) offer interpretable low-rank embeddings.
• Addressing confounding factors is crucial for accurate clustering results.

Purpose of the Study:

• To develop a novel multi-view clustering method, INMTD, for integrating omics and 3D imaging data.
• To derive unconfounded subgroups of individuals by handling unwanted drivers of clustering.
• To demonstrate the method's effectiveness on both synthetic and real-life datasets.

Main Methods:

• Introduced INMTD, a new method combining NMTF and NTD for multi-view clustering.
• Integrated omics and 3D imaging data to generate embeddings.
• Applied techniques to remove confounded embedding vectors for unconfounded clustering.

Main Results:

• INMTD outperformed existing methods on a synthetic dataset, validated by the adjusted Rand index.
• Applied to facial-genomic data, INMTD produced biologically relevant embeddings for individuals, genetics, and facial morphology.
• Unconfounded clustering revealed distinct subgroup characteristics, highlighting genetic and facial features.

Conclusions:

• INMTD effectively integrates omics and 3D imaging data for unconfounded clustering.
• The method provides biologically meaningful interpretations of derived subgroups.
• INMTD facilitates the discovery of distinct individual characteristics from integrated data.