Robust environmental exposure pattern recognition and outlier detection with the open-source R package pcpr

Area of Science:

• Environmental health science
• Epidemiology
• Data science

Background:

• High-dimensional mixture data analysis is crucial in environmental health (EH) for identifying exposure sources.
• Principal Component Pursuit (PCP) is a robust dimensionality reduction technique for pattern recognition in EH.
• Lack of tailored software has limited PCP application in EH research.

Purpose of the Study:

• Introduce pcpr, an open-source R package for accessible Principal Component Pursuit (PCP) in environmental health.
• Provide EH-specific extensions to PCP for enhanced interpretability and data handling.
• Demonstrate the package's utility through a reproducible air pollution source-apportionment analysis.

Main Methods:

• Developed an R package, pcpr, implementing PCP with three EH-specific extensions: non-negativity constraints, missingness accommodation, and detection limit adjustments.
• Applied pcpr to a dataset of 26 PM2.5 constituents in Queens, NYC (2015-2021).
• Conducted a fully reproducible source-apportionment analysis to identify exposure patterns and events.

Main Results:

• The pcpr package successfully extracted four consistent exposure patterns (secondary/tailpipe emissions, traffic, crustal dust, salt) from the PM2.5 mixture.
• Identified 211 outlying exposure events, including notable spikes related to fireworks.
• Demonstrated the package's capability for robust pattern recognition in complex environmental mixtures.

Conclusions:

• The pcpr R package offers a robust, reproducible, and accessible solution for exposure pattern recognition in environmental epidemiology.
• Facilitates the identification of specific pollution sources and extreme events.
• Supports targeted regulatory action by linking identified patterns to potential adverse health outcomes.