Transforming Drug Discovery Using Machine Learning

The speaker discusses the field of digital biology and its potential to shape the next 30 years of science, particularly in the discovery and development of medicines. They explain the concept of end-to-end learning and reducing the dimensionality of the problem to create a new feature or presentation of the domain. The speaker also outlines their work in applying machine learning to multiple steps in the drug development process, including target identification, molecule creation, and patient segmentation for clinical trials.

• Digital biology is the synthesis of data and quantitative biology, enabling the measurement of biology in new ways and scales
• End-to-end learning involves creating a new feature or presentation of the domain by reducing the dimensionality of the problem
• Machine learning is applied to multiple steps in the drug development process, including target identification, molecule creation, and patient segmentation for clinical trials

The speaker gives an example of how end-to-end learning can create a hundred dimensional representation in a million dimensional space, allowing for objects that are far away from each other in the original space to be close by in the manifold. This capability is important in biology discovery, as it allows for the inference of related classes despite not being part of the input.

Daphne Koller, Founder, and CEO of insitro, shares how recent advances in machine learning and medicine enables us to 'rethink' drug discovery and reverse a long-standing trend of declining drug research effectiveness. She also shares how AI can make high-quality clinical predictions that are as good as human beings, if not better, leading to better patient outcomes.