Examining structure-based surrogate selection for quantitative non-targeted analysis.

Area of Science:

• Environmental Chemistry
• Analytical Chemistry
• Mass Spectrometry

Background:

• Quantitative non-targeted analysis (qNTA) is crucial for identifying and quantifying emerging contaminants.
• Current qNTA methods often use surrogate chemicals for calibration, but selection is frequently based on intuition rather than rational criteria.
• This lack of systematic surrogate selection hinders objective assessment and improvement of qNTA models.

Purpose of the Study:

• To systematically evaluate the impact of chemical structure on surrogate selection for qNTA.
• To develop and compare structure-based surrogate selection strategies against random selection.
• To introduce a metric for quantifying surrogate coverage in chemical space.

Main Methods:

• Calculated chemical space embeddings using LC-HRMS data and 2D molecular descriptors.
• Determined analyte leverage within the chemical space using EPA's ENTACT dataset.
• Implemented and compared structure-based and random surrogate selection strategies using qNTA metrics.
• Proposed and examined the "leveraged averaged representative distance" (LARD) metric.

Main Results:

• Structure-based surrogate selection strategies can enhance qNTA model performance.
• A sufficiently large random surrogate sample can achieve comparable accuracy to smaller, chemically informed sets.
• The LARD metric effectively quantifies surrogate coverage within the chemical space.

Conclusions:

• Rational, structure-informed surrogate selection offers benefits for qNTA accuracy and reliability.
• While informed selection is advantageous, random sampling can be effective if sample size is adequate.
• Findings guide researchers in optimizing surrogate selection for robust qNTA.