TEMPTED: time-informed dimensionality reduction for longitudinal microbiome studies.

Area of Science:

• Microbiome research
• Computational biology
• Longitudinal data analysis

Background:

• Understanding microbiome dynamics over time is essential for health research.
• Existing methods struggle with high-dimensional longitudinal data and varying sampling rates.
• Characterizing temporal microbial shifts requires advanced analytical approaches.

Purpose of the Study:

• Introduce TEMPoral TEnsor Decomposition (TEMPTED), a novel time-informed dimensionality reduction technique.
• To effectively analyze high-dimensional longitudinal microbiome data, treating time as a continuous variable.
• To improve the characterization of microbial dynamics, beta-diversity analysis, and data reproducibility.

Main Methods:

• TEMPTED employs tensor decomposition to model time-series microbiome data.
• It handles varying temporal sampling intervals and captures continuous temporal information.
• Learned representations are transferable to new datasets for enhanced reproducibility.

Main Results:

• TEMPTED achieved 90% accuracy in phenotype classification in simulations, outperforming existing methods.
• Identified key vaginal microbial markers associated with term and preterm births in real-world data.
• Demonstrated robust performance across diverse datasets and sequencing platforms.

Conclusions:

• TEMPTED is a powerful tool for analyzing longitudinal microbiome data.
• The method accurately captures temporal microbial dynamics and aids in identifying health-related biomarkers.
• TEMPTED enhances reproducibility and performance in microbiome research across different data types.