Join us on Tuesday, January 26, 2021 - 12:00 PM to 1:30 PM EST
We spend a lot of time building models and comparing them to other models. The field generally agrees that forward-looking predictions are the best validation, but even prospective validation can be misleading if it does not consider the actual deployment context: How does the model affect compound selection? What controls are in place to ensure that downstream uses of model predictions are consistent? As a concrete example, I'll review our recent work with DNA-encoded libraries. I'll conclude with a summary of the Open Reaction Database, a large-scale data gathering effort with some interesting possibilities for prospective applications.