Computer Vision to analyze chemical signatures: a novel workflow for rationalizing raw data exploration in GC×GC

This work delves into Computer Vision in the context of Artificial Intelligence and its ability to extract meaningful information from digital images. This concept is particularly relevant to the field of AI, where the processing of large amounts of data and the ability to derive insights from it is of utmost importance [1]. One of the techniques where AI can be applied is in combination with comprehensive two-dimensional gas chromatography (GC×GC), which provides highly detailed information on the chemical composition of a sample generating chromatogram images with a multidimensional array of data. However, the high amount of these data can make it difficult to interpret and analyze information encrypted. This is where AI techniques such as Computer Vision come into play, as they can help rationalize raw data exploration, leading to an understanding of the biological phenomena related to specific chemical signatures and molecular patterns. In this work, a new workflow for Computer Vision based on pattern recognition algorithms, such as combined untargeted and targeted (UT) fingerprinting is presented. This workflow involves several steps, starting with the generation of composite class images for representative samples' classes. These images are then processed to create a feature template with reliable peaks and peak-regions. The feature template is then pruned to include targeted components while eliminating bleeding peaks and artifacts. Once the feature template is finalized, it is applied to all samples' images and composite class images. Finally, pair-wise comparisons are made to highlight quantitative pattern differences and link them to the chemistry of targeted compounds and tracked features of untargeted compounds. To demonstrate the effectiveness of this workflow a sample set from a research project on artisanal butter was explicated. The sample set examines the changes in volatile patterns during the production process, from raw sweet cream to ripened butter. The application of the workflow to this sample set shows how the evolution of the volatile fraction along the production chain can be captured, providing valuable insights into the chemical composition and quality changes. Overall, the importance of Computer Vision in the field of GC×GC was highlighted, particularly when processing large amounts of data to extract meaningful information from it. The workflow described in the contribution provides a structured approach to rationalizing raw data exploration, leading to a better understanding of complex phenomena such as the chemical composition of samples and their changes over time.