Developing multiple drugs of abuse testing (DAT) assays in parallel for the cobas^® Mass Spectrometry system demands an efficient optimization workflow. This talk details our approach to optimizing the HPLC gradient, a typical challenge with a vast factor space in LCMS method development.

We began with a constrained, I-optimal designed experiment of over 80 runs to generate the rich data set needed for robust modeling. The core of the presentation focuses on the classical analysis, walking through the process of building predictive models for many responses and integrating them into a unified optimization framework to identify robust operating conditions.

In parallel, we explored a Bayesian learning approach. This dual strategy was designed not only to optimize the current assays but also to learn which modeling philosophy is better suited for future projects. This talk offers a unique comparison of these two workflows, providing insights into their practical application and the final, experimentally verified results. Attendees learn the relative strengths and trade-offs of a classical DOE approach versus Bayesian learning methods in an industrial R&D setting.