How to predict across studies: Massively increasing the numbers of antibody-virus interactions using machine learning

  • Starts: 2:30 pm on Friday, September 23, 2022
  • Ends: 3:30 pm on Friday, September 23, 2022
A central challenge in every field of science is to use the information gained from previous studies to predict the outcomes of future experiments. Here, we consider antibody responses against the rapidly-evolving influenza virus, where we seek a broad response that inhibits the hundreds of currently-circulating viruses, thousands of previously-circulating strains, and all future variants. Yet most studies only examine several dozen viruses, making it difficult to combine datasets or discern underlying principles. Rather than viewing this as a weakness, we instead leverage patterns across datasets to predict how an antibody response from one study would behave against any virus from any other study. This approach predicts millions of new interactions, and it calls for a shift in mindset when analyzing data from “what you see is what you get” into “what anyone sees is what everyone gets.”
LSE 103
Tal Einav,
Fred Hutch Cancer Center