Measured vs predicted: from molecules to patients

​Measured vs predicted: from molecules to patients

​At our first evening “Bio x Engineering”, most tables ended up somewhere similar: we have a number we can measure and an outcome we care about, and we are not sure how well one tells us about the other. This time we want to go deeper into this.

The evening has two parts. The first is about the data going into our models: what do we need to record about an experiment to know that a model has learned the biology and not the batch effects? The second is about the proxies coming out: how do we test whether a readout predicts anything about a patient, and can one field borrow another’s way of doing it?

By the end we would like to have two things written down, even if they are rough. A first list of what should be recorded about an experiment or an AI output for the data to be reusable by someone else, beyond what is already common practice in the field. And a sketch of a framework for testing a proxy against real outcomes, one that could be adapted to different diseases, fields and applications.

​We want people from different backgrounds in the room so that the discussion goes somewhere, and so that what comes out of it is something we can share with the community and, hopefully, use to build better solutions in biotech and medtech.

​Bring an example from your own work where the measurement and the outcome did not agree, and think about what you would have needed to know to catch it earlier.