DMGAN: Bridging AI and Chemistry with Enhanced GC-MS Data Generation

Data from gas chromatography-mass spectrometry (GC-MS) represent significant deviations between the large peak values of a particular substance and the small peak values of other components. Such deviations have many limitations for applying traditional deep learning techniques. Our proposed Dynamic Modulation Generative Adversarial Network (DMGAN) provides a solution to these limitations and generates a variety of synthetic data that enable real experiments to be efficient. We compare it with various models using pearson correlations, a representative metric that measures the similarity of chromatographic data, with DMGAN showing a quality exceeding 0.99. Furthermore, we show how it can be used for models in the existing AI-chemistry field using the generated synthetic data, showing a performance improvement of over 50%. This paper highlights the possibility that DMGAN can establish a strong connection between AI and chemistry.