Novel Application of Survival Models for Predicting Microbial Community Transitions with Variable Selection for...

Area of Science:

• Microbiome research
• Statistical modeling
• Environmental science

Background:

• Survival analysis is crucial in medicine but underused in microbiome research due to data complexity.
• Predicting microbial community events requires robust methods for high-dimensional, sparse data.

Purpose of the Study:

• To adapt and enhance Cox proportional hazards (Cox PH) survival models for microbiome data.
• To develop and validate a novel feature selection method (SuRFCox) for identifying microbial biomarkers.

Main Methods:

• Applied Cox PH survival models to environmental DNA (eDNA) data.
• Developed SuRFCox for selecting relevant taxonomic variables, outperforming existing methods.
• Compared selection methods using simulations and real-world data for forecasting harmful cyanobacterial blooms.

Main Results:

• SuRFCox demonstrated superior performance in identifying biomarkers and predicting events.
• Cox PH models with SuRFCox predictors were robust to varied data conditions.
• Accurate and consistent prediction of harmful cyanobacterial blooms was achieved over multiple seasons.

Conclusions:

• SuRFCox provides a robust and accurate method for biomarker discovery in microbiome survival analysis.
• This approach enhances risk assessment for microbial community-mediated events, including ecological and public health impacts.
• The findings advance theoretical and practical understanding of critical community transitions.