Weighted fusion regularisation and predicting microbial interactions with vector autoregressive model

Area of Science:

• Statistics
• Bioinformatics
• Time Series Analysis

Background:

• Multivariate autoregressive (MVAR) models are widely used for analyzing time series data.
• Handling correlated variables in MVAR models remains a challenge.
• Existing regularization methods may not fully account for variable correlations.

Purpose of the Study:

• To develop a novel regularization method for MVAR models that incorporates variable correlations.
• To theoretically analyze the grouping effect of the proposed weighted fusion regularization.
• To apply and validate the method on real-world time series data, including human gut microbiome data.

Main Methods:

• Developed a weighted fusion regularization technique for MVAR models.
• Utilized probability methods to demonstrate the grouping effect on correlated predictors.
• Quantitatively estimated coefficient differences for highly correlated variables.
• Applied the method to diverse time series datasets, focusing on microbiome data.

Main Results:

• The weighted fusion regularization exhibits a grouping effect on linear models.
• Demonstrated that coefficients of highly correlated predictors have small, quantitatively estimated differences.
• The proposed method showed superior performance compared to other VAR-based models.
• Successfully extracted relevant microbial interactions from gut microbiome time series data.

Conclusions:

• The novel weighted fusion regularization method effectively addresses correlations in MVAR models.
• The approach offers improved performance and insights into complex time series data.
• This method has significant potential for applications in fields like microbiome research.