Abstract

Grand challenges in biology and medicine often lack annotated examples and require generalization to entirely new scenarios not seen during training. However, standard supervised learning is incredibly limited in scenarios, such as designing novel medicines, modeling emerging pathogens, and treating rare diseases. In this talk, I present our efforts to overcome these obstacles by infusing structure and knowledge into learning algorithms. First, I outline our subgraph neural networks that disentangle distinct aspects of subgraph structure in networks. I will then present a general-purpose approach for few-shot learning on graphs. At the core is the notion of local subgraphs that transfer knowledge from one task to another, even when only a handful of labeled examples are available. This principle is theoretically justified as we show the evidence for predictions can be found in subgraphs surrounding the targets. I will conclude with applications in drug development and precision medicine where the algorithmic predictions were validated in human cells and led to the discovery of a new class of drugs.