Biological molecules, such as proteins, are highly diverse and hold great therapeutic potential. However, the discovery of functional biological molecules can be challenging and costly due to their complex design space and the need to consider various design criteria for safety and efficacy. In this talk, I will discuss how machine learning and computational methods can be used to accelerate the biomedical discovery cycle in a holistic fashion. I will first motivate an AI-driven paradigm that combines interpretable and uncertainty-aware deep learning models with active learning, black-box optimization techniques to efficiently guide iterative experiment design and refinement. I will then introduce a novel approach for training deep ensemble models that produce reliable out-of-distribution prediction and better uncertainty estimates for efficient Bayesian Optimization over biological sequence space. To showcase the effectiveness of my approaches, I will present two high-impact applications in real-world therapeutics design. Firstly, I will present a pioneering deep-learning-driven antibody design framework that transforms conventional screening-based discovery by learning from high-throughput phage display data and proposing superior candidates through ML-directed optimization. My method produces monoclonal antibodies with experimentally verified better affinity and specificity without structural information. Secondly, I will introduce a novel COVID-19 peptide vaccine formulation based on ML and combinatorial optimization. Our vaccine is designed to provide durable, pan-variant, and broadly effective T-cell response and rectifies the insufficient coverage of underrepresented groups. Our designed mRNA vaccine was able to prevent mortality in transgenic mice infected with the Beta variant, and showed improved T-cell response and lower side effect compared to the Pfizer-BioNTech vaccine.
Ge Liu is an applied scientist at AWS AI Labs. She received her Ph.D. from MIT EECS department, advised by Professor David Gifford. Her research lies in the intersection of machine learning and biology, with a special interest in therapeutic biological molecule design. She develops reliable, interpretable, and efficient machine learning and computational techniques for solving high-stake problems in synthetic biology, immunology, and functional genomics, with a long-term mission of accelerating biomedicine discovery cycles with ML. The principled methods she developed can be widely used in high-impact applications such as antibody drug discovery, cancer immunotherapy, and vaccine development while also applicable to domains such as sequential modeling, recommender systems, and active learning. She is the recipient of the David S. Y. Wong Fellowship at MIT, and her Ph.D. thesis won the MIT EECS George M. Sprowls Ph.D. Thesis Award in AI and Decision-Making.