Exploring the Feasibility of Using Generative AI and Signal Processing in Solving Separation Science Problems

Contrary to widespread belief, artificial intelligence (AI) has been utilized in the chemical sciences since the 1960s, perhaps originating from concepts developed in the Stanford Artificial Intelligence Project. The recent surge in AI-related publications over the past three years is also reshaping how analytical chemistry is researched and its data processed. In today's world, the breadth of knowledge required to claim expertise in multiple fields has expanded beyond the realistic capacity of an individual scientist. What does the generative AI have in store for chromatographers and analytical chemists? This talk explores the feasibility of using generative AI as a highly knowledgeable "colleague" for solving chromatographic problems, particularly those involving advanced signal processing techniques such as denoising, peak width reduction by deconvolution, and decomposing overlapping peaks. For example, noise and instrumental distortions affect recorded chromatograms in many negative ways. This presentation covers the possibility of reconstructing an estimated trace of the "true" chromatogram free from detection artifacts and noise corruption. The first part discusses new advanced denoising techniques that leverage mathematical optimization and how ChatGPT can help in implementation. Traditional digital filters embedded in current chromatographic instruments often reduce noise but at the cost of broadening chromatographic peaks. New techniques will be introduced to keep the peak information intact during the denoising process. The second part focuses on enhancing the digital resolution of denoised chromatograms using regularization deconvolution methods and iterative curve fitting techniques.