Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore the critical transition from predictive machine learning to causal inference in biomedical applications through this conference talk from the EWSC-MIT EECS Joint Colloquium Series. Learn how Stanford University Professor Emily Fox addresses the fundamental challenge that predictive accuracy alone is insufficient for decision-making in drug discovery and health monitoring, emphasizing the necessity of establishing causality. Discover insitro's innovative approach to therapeutic target discovery that triangulates clinical and cellular perspectives, utilizing human genetics as natural experiments to identify causal genetic drivers while leveraging ML-derived precision phenotypes from multimodal clinical data. Examine how perturbation screens in disease-relevant models measure phenotypic consequences across high-content modalities including microscopy and omics data. Understand the concept of cross-lens concordance between ClinML and CellML approaches and how it builds confidence in causal target identification. Investigate key sources of bias in machine learning models and explore modeling strategies including multimodal embeddings and sparse autoencoders for aligning and interrogating ML-derived phenotypes. Delve into the Stanford research on incorporating mechanistic knowledge as causal inductive bias to address the problem where high predictive accuracy can coexist with low causal validity in observational data. Learn about the development of hybrid models that combine mechanistic ODE dynamics with flexible neural network components, trained using a novel hybrid loss function that balances predictive performance with causal validity. See practical applications through a type 1 diabetes case study demonstrating state-of-the-art performance in modeling glucose dynamics post-exercise while maintaining causal interpretability.
