A Data‑Driven, Post‑Acquisition Quality Diagnostic Pipeline for Isotope Analysis by MC-ICP-MS

Abstract

The post acquisition evaluation of data from multi collector inductively coupled plasma mass spectrometry (MC ICP MS) is a specialized but subjective process, relying on expert judgment to diagnose measurement quality. This lack of standardization can hinder reproducibility and the development of autonomous analytical workflows. To address this, we developed a generalizable, data driven diagnostic pipeline that formalizes expert judgment into an automated, machine learning (ML) assisted framework. The pipeline is demonstrated using mercury (Hg) isotope analysis as a representative and demanding case, owing to its sensitivity to instrumental instability from cold vapor introduction. It integrates three core modules: (1) robotic process automation for batch data extraction from vendor software, (2) automated calculation of final isotope ratios (δ and Δ values), and (3) a hierarchical diagnostic system that formalizes expert judgment into an automated, auditable process. This system first applies literature derived expert rules, then employs interpretable ML models, trained on 26,218 historical Hg measurements, to perform objective quality triage. A binary classifier flags abnormal measurements based on δ202Hg, Δ199Hg, Δ200Hg, and Δ201Hg with a test macro F1 score of 0.996 (AUC ≈ 1.0). Subsequently, a cost sensitive multi class model diagnoses the most probable cause of failure (e.g., instrumental instability). External validation confirms the models replicate expert decisions with high fidelity, demonstrating that they effectively learn from expert‑labeled data. While validated for Hg, the framework’s modular design supports adaptation to other isotope systems by replacing element specific components; any new application requires dedicated validation with its own dataset and element-specific expert rules. By unifying automation, explicit rules, and interpretable ML, this work establishes a foundational software method for intelligent, post acquisition quality assurance, providing a scalable template to standardize data assessment across MC ICP MS laboratories and a critical step toward fully autonomous analytical workflows.