Enabling network inference methods to handle missing data and outliers.

Area of Science:

• Multidisciplinary network inference
• Data analysis in biological sciences, chemistry, economics, and sociology

Background:

• Complex network inference from data is crucial but challenged by data quality issues.
• Existing methodologies often fail to address missing data or outliers effectively.
• Proper handling of incomplete and erroneous data is essential for reliable network inference.

Purpose of the Study:

• To introduce a novel approach for handling missing data and detecting/correcting outliers in datasets.
• To enhance the capability of network inference methods to analyze incomplete and faulty datasets.
• To provide a robust data curation step for network inference.

Main Methods:

• Development of Trimmed Scores Regression (TSR) utilizing multivariate projection to latent structures.
• TSR imputes missing values coherently with the latent data structure.
• TSR detects and corrects outlier values through robust estimation.

Main Results:

• TSR enables network inference on incomplete datasets by imputing missing values.
• The method effectively substitutes erroneous data points with accurate estimations.
• Demonstrated integration of TSR with the MIDER network inference method.

Conclusions:

• The TSR methodology significantly expands the scope of network inference to include previously unmanageable datasets.
• Comparative studies confirm TSR's superior performance over alternative missing data imputation techniques.
• TSR offers a comprehensive solution for both missing data and outlier issues in network analysis.