IC/SV Colloquium: Machine Learning-Guided Treatment Discovery and Planning
By: Charlotte Bunne - ETH Zurich
IC/SV Faculty candidate
Abstract
In recent years, massively parallel high-throughput methods have changed the course of modern drug discovery. While providing us with an unprecedented resolution into molecular processes, they require scalable and principled algorithms that integrate the most recent insights from human biology, are well aligned with the constrained nature of experiments, and incorporate the inherent structure of macromolecules and tissues. These criteria have guided my research toward mathematically grounded deep learning solutions, using notably optimal transport and geometric modeling. In this seminar, I will demonstrate how integrating these principles into the design of learning algorithms shifts our ability to predict heterogeneous patient treatment responses to the single-cell level, model combination therapies, and trace developmental differentiation processes. These novel deep learning approaches not only achieve state-of-the-art quantitative improvements over prior works but also open new frontiers in a current large-scale clinical study to predict treatment responses of unseen patients. Altogether, my work on neural optimal transport and geometric deep learning shows that innovations in the design of machine learning algorithms will be crucial for accelerating the discovery of therapeutics and proposing personalized treatment plans to patients.
Bio
Charlotte Bunne is a PhD student in Computer Science at ETH Zurich working with Andreas Krause and Marco Cuturi and currently a visiting researcher at the Broad Institute of MIT and Harvard hosted by Anne Carpenter and Shantanu Singh. Before, she worked with Stefanie Jegelka as a Master student at MIT. With a focus on accelerating therapeutics discovery and proposing personalized treatment plans for patients, she develops mathematically grounded deep learning solutions based on optimal transport and geometrical modeling. The neural optimal transport methods she innovates prove successful within both the medical and the machine learning community: These tools are the current state-of-the-art in predicting dynamic single-cell responses of a patient and are a key ingredient of an ongoing clinical cohort study while at the same time receiving best paper awards at machine learning conference workshops at NeurIPS’18, ICML’20, and ICML’21. Throughout her studies, Charlotte has been a Fellow of the German National Academic Foundation. She is a recipient of the ETH Medal.
More information
IC/SV Faculty candidate
Abstract
In recent years, massively parallel high-throughput methods have changed the course of modern drug discovery. While providing us with an unprecedented resolution into molecular processes, they require scalable and principled algorithms that integrate the most recent insights from human biology, are well aligned with the constrained nature of experiments, and incorporate the inherent structure of macromolecules and tissues. These criteria have guided my research toward mathematically grounded deep learning solutions, using notably optimal transport and geometric modeling. In this seminar, I will demonstrate how integrating these principles into the design of learning algorithms shifts our ability to predict heterogeneous patient treatment responses to the single-cell level, model combination therapies, and trace developmental differentiation processes. These novel deep learning approaches not only achieve state-of-the-art quantitative improvements over prior works but also open new frontiers in a current large-scale clinical study to predict treatment responses of unseen patients. Altogether, my work on neural optimal transport and geometric deep learning shows that innovations in the design of machine learning algorithms will be crucial for accelerating the discovery of therapeutics and proposing personalized treatment plans to patients.
Bio
Charlotte Bunne is a PhD student in Computer Science at ETH Zurich working with Andreas Krause and Marco Cuturi and currently a visiting researcher at the Broad Institute of MIT and Harvard hosted by Anne Carpenter and Shantanu Singh. Before, she worked with Stefanie Jegelka as a Master student at MIT. With a focus on accelerating therapeutics discovery and proposing personalized treatment plans for patients, she develops mathematically grounded deep learning solutions based on optimal transport and geometrical modeling. The neural optimal transport methods she innovates prove successful within both the medical and the machine learning community: These tools are the current state-of-the-art in predicting dynamic single-cell responses of a patient and are a key ingredient of an ongoing clinical cohort study while at the same time receiving best paper awards at machine learning conference workshops at NeurIPS’18, ICML’20, and ICML’21. Throughout her studies, Charlotte has been a Fellow of the German National Academic Foundation. She is a recipient of the ETH Medal.
More information
Practical information
- General public
- Free
Contact
- Host: Martin Jaggi