Correlating High-dimensional longitudinal microbial features with time-varying outcomes with FLORAL

Area of Science:

• Microbiome research
• Statistical modeling
• Biomarker discovery

Background:

• Longitudinal microbiome studies require methods to correlate time-dependent patient data with microbial samples.
• Current methods often apply a single modeling approach across all features, followed by false discovery rate (FDR) adjustment.
• This can limit sensitivity and the ability to identify true associations.

Purpose of the Study:

• To develop an advanced statistical strategy for analyzing longitudinal microbiome data.
• To identify microbial biomarkers associated with time-varying patient characteristics.
• To improve sensitivity and FDR control in microbiome biomarker discovery.

Main Methods:

• Utilized log-ratio penalized generalized estimating equations for direct modeling.
• Treated microbial features as high-dimensional compositional covariates.
• Employed cross-validation for variable and model selection, including correlation structures.

Main Results:

• The proposed method demonstrated superior sensitivity compared to existing state-of-the-art approaches.
• Achieved robust control of the false discovery rate (FDR).
• Successfully identified gut health indicators and relevant microbial markers in cancer patient data.

Conclusions:

• The novel method offers enhanced capabilities for biomarker discovery in longitudinal microbiome studies.
• Demonstrated robust utility in real-world applications, such as analyzing dietary intake and gut microbiota.
• The method is available via the open-source R package FLORAL.