Phil Kay and Chandramouli Ramnarayanan, JMP
Advances in AI, lab automation and closed-loop optimization promise big productivity gains for pharma R&D. But experimenting with maximum efficiency - that is, the least number of runs - will always be important. This is especially true where there is an ethical imperative, such as in pre-clinical animal experiments. Recent advances in statistical Design Of Experiments (DOE) including Definitive Screening Designs (DSDs) and the broader class of Orthogonal Minimally Aliasing Response Surface (OMARS) designs have given us new options for understanding complex systems with small experiments. Self-Validated Ensemble Modelling (SVEM) is an innovative analysis approach that applies ideas from Machine Learning to small data from designed experiments. In this presentation we will show how SVEM works and how it can overcome common challenges when you are building useful models of complex systems from small experiments.