What kind of model will strike the best balance of fit and parsimony for your PK data? A one compartment? Two or three? What about elimination? Is it linear, Michaelis-Menten, or TMDD?
These are just two of the many structural considerations you face when building a popPK model. While known drug and subject characteristics may help you fix some structures, it’s often convention or bias that determine the rest. The greater the role that bias plays, the wider the gulf between “the best fit” and “the best fit that you happen to find.”
Machine learning offers a powerful corrective. Algorithms quickly sample models across a vast model space (e.g. every combination of structural features you wish to consider). Over time, they concentrate their search in the most promising areas, exploiting best fits while still exploring “second bests”, to ultimately converge on the optimal model. Whereas an exhaustive search of all possible models could take weeks or months of computation, an ML-driven approach can shorten the time to hours.
Join three of Certara’s expert modelers for a gentle introduction to machine learning in our field. They’ll begin with a helpful analogy to orient you to model search spaces and methods for converging on the model type best suited to your data. From there, they will discuss the benefits to research and clinical practice that an ML-driven approach can yield. Finally, our team will illustrate some of the tools, like pyDarwin and Darwin in Pirana, that turn theory into a clear and learnable practice, including an intuitive GUI that makes getting started easy.
講演者
- Mark Sale, MD, Vice President, Integrated Drug Development, Certara
- Keith Nieforth, PharmD, Senior Director, Pharmacometric Software, Certara
- James Craig, Staff Software Engineer, Certara