Dimensionality reduction of longitudinal 'omics data using modern tensor factorizations.

Area of Science:

• Biomedical data analysis
• Computational biology
• Precision medicine

Background:

• Longitudinal omics data are crucial for precision medicine, capturing individual variations.
• Biomedical studies often generate high-dimensional, sample-limited datasets, posing analytical challenges.
• Conventional methods struggle to filter noise and identify true biological signals in complex omics data.

Purpose of the Study:

• To introduce tcam, a new unsupervised tensor factorization method for multiway data analysis.
• To address the challenges of analyzing feature-rich and sample-limited longitudinal omics data.
• To provide a tool for uncovering biologically relevant signals beyond inter-individual variation.

Main Methods:

• Developed tcam, an unsupervised tensor factorization method based on tensor-tensor algebra.
• Characterized tcam's mathematical properties, including data trait preservation and out-of-sample extension.
• Applied tcam to re-analyze real-world human experimental longitudinal omics datasets.

Main Results:

• Tcam preserves geometric and statistical data traits, enabling discovery beyond inter-individual variation.
• The method offers natural out-of-sample extension, facilitating integration into machine learning workflows.
• Re-analyses confirmed tcam's utility and theoretical properties in longitudinal omics data analysis.

Conclusions:

• Tcam is a powerful new tool for analyzing complex longitudinal omics data in precision medicine.
• The method enhances the interpretation of high-dimensional biological datasets by filtering noise effectively.
• Tcam's capabilities support the identification of subtle biological signals crucial for personalized healthcare.