AI-based Denoising of LC-MS data via Generative Adversarial Networks: A Proof of Concept

CASSS: Mass Spec, Rockville, MD, USA
September 10, 2024

Denoising parameterization traditionally requires expert understanding and monitoring, and presents a source of variability. Here, we present a novel, AI-based, pre-processing method for denoising Liquid Chromatography-Mass Spectrometry (LC-MS) data using conditional Generative Adversarial Networks (cGANs). This eliminates the need for any parameterization of the denoising method.
The new method achieves comparable performance vs. state-of-the-art methods, especially by scoring on downstream results, e.g., regarding coverage of Heavy Chain (HC) and Light Chain (LC) or the critical quality attributes (CQAs) in LC-MS-based MAM data analysis.
This workflow demonstrates robustness across diverse data sets and doesn’t require parameterization or other tuning. It helps to further automate data analysis, and enhancing efficiency and reliability. This innovative approach highlights the potential of AI assisted analysis for complex LC-MS data.