AI Summary & Key Takeaways
AI InsightCorrecting these limitations is essential to implement high-performance AI models in a more realistic, transparent, and transferable analytical tool for modern analytical chemistry, and this is of great importance for their translation and successful application in the field of analytical chemistry.
📌 3 Key Takeaways
Abstract
Artificial intelligence (AI) is transforming analytical chemistry by making it even more efficient to process and interpret complex high-dimensional analytical data. Machine learning (ML) and deep learning (DL) approaches have huge potential in spectroscopy, chromatography, mass spectrometry, and other analytical applications and thus enhance classification, prediction, method optimization, and data interpretation. In this review we examine the current state of AI applications in analytical chemistry, especially for the commonly employed approaches (PLS, SVM, ANN, random forests, and CNN) and their features and limitations. In addition to predictive accuracy, the main challenges to reliable implementation are overfitting and dataset size and quality, preprocessing dependence, model complexity and interpretability, external validation, and transferability between different instruments and laboratories. To date, no single approach is more efficient, and hence, the selection of a model is based on the characteristics of analytical data used, application of the data, and validation strategy. The major knowledge gaps in this regard are: a lack of standardized datasets, limited and non-standardized external and inter-laboratory validation of models, and a lack of chemically meaningful model interpretation. Correcting these limitations is essential to implement high-performance AI models in a more realistic, transparent, and transferable analytical tool for modern analytical chemistry, and this is of great importance for their translation and successful application in the field of analytical chemistry.