Data Science Tools for Advanced Method Development and Prediction in Analytical Measurements

With increasing complexity for the demands of chemical measurements, new analytical techniques must be developed. Method development is a time-consuming process, which is further hampered when a plethora of variables require optimization. Advanced data techniques based on machine learning and surrogate optimization hold promise for reducing method development time. In the context of on-line supercritical fluid extraction - supercritical fluid chromatography (SFE-SFC), we have applied surrogate optimization for method development. This approach lessens the number of steps needed to arrive at acceptably optimal analysis conditions and can handle a larger number of variables compared to traditional factorial design and response surface methodologies. Surrogate optimization is also flexible in terms of types of inputs. One promising approach for encoding variables into the design is to use molecular feature generation, where chemical structures are assessed based on their atom connectivity rather than other calculable properties or measured values. The use of molecular feature generation has been investigated in the context of predicting vacuum ultraviolet absorption spectroscopy, and its use for SFE-SC is being explored. Composite objectives can also be constructed to more fully assess analytical performance. The goal of this talk is to convey some data science tools that have been adopted and are currently being developed to advance analytical measurement science.