New Workflow for Rationalizing Raw Data Exploration in GC×GC: Computer Vision to Unravel Diagnostic Signatures

Introduction Computer vision is a branch of artificial intelligence (AI) that allows systems to extract useful information from digital images and make predictions based on that information [1]. Chromatographic fingerprinting on patterns obtained through comprehensive two-dimensional chromatography (C2DC) aims to detect, align, and compare features in analyzed samples. By representing C2DC chromatograms as digital images, the process enables higher-level information about unique composition and its relation to the studied phenomenon. Chromatographic images can be processed using different feature types, such as peak features, to profile components across multiple samples. The process of aligning and comparing 2D peak patterns is known as chromatographic fingerprinting, which can also be considered a Computer Vision approach. Materials and Methods In this work we focused on the processing of analytical data; in particular on the interpretation of two-dimensional chromatograms to understand the evolution of the volatilome of a complex during the production chain. In this field AI techniques such as Computer Vision can help rationalize raw data exploration, leading to an understanding of the biological phenomena related to specific chemical signatures and molecular patterns. A sample set from a study project on artisanal butter was explained to show the efficacy of this procedure. The sample set looks at how volatile patterns vary from raw sweet cream to ripened butter during the production process. Results and Discussion This work presents a new approach for computer vision that is based on pattern recognition techniques such combined untargeted and targeted (UT) fingerprinting. This procedure entails a number of steps, beginning with the creation of composite class pictures for representative samples' classes. Then, a feature template with reliable peaks and peak-regions is created using these images. After that, bleeding peaks and artefacts are removed from the feature template while keeping the targeted components. All sample images and composite class images are subjected to the feature template once it has been finalized. In order to highlight quantitative pattern differences and relate them to the chemistry of targeted compounds and monitored features of untargeted compounds, pair-wise comparisons are then performed. Figure 1. Comparative overlaid visualization between raw sweet cream and ripened butter. Different colorization emphasizes differences between class images. The workflow's application to this sample set shows how the evolution of the volatile fraction can be tracked throughout the production chain, giving important insights into changes in chemical composition and quality. In figure 1 through the use of Computer vision it is easy to appreciate the compositional difference at the volatilome level between two production steps: raw sweet cream and 40 days ripened butter. Conclusions To capture the significant changes in food volatilome throughout processing, GC×GC-TOFMS with thermal modulator proved to be highly successful. Identification of diagnostic patterns of process and quality markers may be hindered by the interaction of many diverse variables. Computer Vision allows an effective comparative visualization of Class Images realized by realigning and combining 2D chromatograms from a Samples' Class. Pair-wise differences are conveniently recorded, and UT fingerprinting makes it simple to obtain higher level data. Investigating at the molecular level is straightforward and reliable. References 1) https://www.ibm.com/topics/computer-vision