Abstract

With the costs of drug discovery and development continuing to rise and the low hanging fruit seemingly diminished, many organizations are moving to machine learning-based strategies for development. One limitation of this strategy is that the predictive models learned during these processes are only as good as the data used in training. Using laboratory automation, one can generate relatively large datasets quickly and reproducibly. However, large datasets may not actually contain enough informative data to learn accurate predictive models. In order address this problem for complex systems, artificial intelligence driven closed loop experimental processes can be used to choose which experiments to run next to maximize information gain relative to experimental expenditures. This yields more accurate predictive models with lower experimental expenditures. At Carnegie Mellon University, we started the first Automated Science Master’s Degree program specifically designed to train students to develop these sorts of closed loop experimental processes. There are three major educational components covered in the program. First, students in the Automation Lab learn how to generate biological data on automated equipment. Second, they learn to analyze those data using computational biology and machine learning strategies. Third, they learn to drive the automated equipment using artificial intelligence to iteratively select informative experiments. This talk will be focused on the educational aspects of the program directed toward drug discovery.