# Automated Near Real‐Time QC for LC‐HRMS

## TL;DR

This paper introduces an automated system for real-time quality control in LC-HRMS measurements, helping to detect and address issues as they happen.

## Contribution

A modular MATLAB workflow for near real-time QC in LC-HRMS, combining multivariate statistical process control and immediate alerting.

## Key findings

- The system successfully flagged 25/25 injections during a power supply failure and 17/25 during a pump malfunction.
- MSPC detected mass error anomalies even when univariate limits were not breached during an AC outage.
- MSPC and univariate thresholds showed strong agreement in flagging problematic samples.

## Abstract

The quality of analytical measurements is typically evaluated after completion of the entire, or possibly multiple, measurement batch(es). Automated, near real‐time quality control (QC) during LC‐HRMS acquisition can prevent reruns and sample loss by flagging issues as they occur. Functionality was evaluated by retrospective application to 5 years of river‐water surveillance.

We present a modular MATLAB workflow that tracks isotopically labelled internal standards for peak height, retention time and mass error against rolling, method‐specific expectations; applies multivariate statistical process control (MSPC; PCA with Hotelling's T
2 and SPE on intensity/retention time ratios and mass error); issues immediate email alerts; and logs outcomes to a PostgreSQL database/Grafana dashboard for trend analysis. Also, qualitative target screening via cosine‐similarity MS2 checks against a local library, retention time correction, robust peak‐height/noise estimation, configurable limits and automated vendor‐to‐open format conversion.

In a high‐voltage power‐supply failure, 25/25 injections were flagged due to abnormal intensity patterns; during an organic‐pump malfunction, 17/25 were flagged for retention drift up to and beyond the extraction window; and during an air‐conditioning (AC) outage, MSPC detected mass error anomalies even when the ±10 ppm univariate limit was not breached. MSPC closely agreed with univariate thresholds: 95.7% of samples flagged by univariate rules were also flagged by MSPC (≈4.3% Type II), while 92.5% of MSPC‐flagged samples violated at least one univariate rule (≈7.5% Type I).

These capabilities enable immediate detection, triage and documentation of performance excursions, support proactive maintenance (e.g., column aging or pump delivery issues), minimise downtime and safeguard precious samples. Although showcased on a specific LC‐HRMS setup and matrix, the workflow is instrument‐agnostic and broadly applicable to internal‐standardised LC‐HRMS methods.