Why did the first AI approach to designing cancer switches fail, and how was it fixed?

Question

Accepted Answer

_What works on lab cell lines can collapse on real patient samples, and the fix was to change the question._

> The original concept used a switch that nature provides, an endogenous promoter. It looked really good in cell lines that never change in the lab, but the moment we tried it on individual real cancer patient samples from the operating room, the whole thing failed completely. So instead of interrogating every single mutation, we asked the opposite question and looked at the downstream commonalities, the hallmarks of cancer, and identified master transcription factor binding sites that are dysregulated.

Source: [Tomorrow's Medicine, with Cyriac Roeding (eCornell Keynote)](https://ecornell.cornell.edu/keynotes/view/K052126/)

Q&A, machine learning

Why did the first AI approach to designing cancer switches fail, and how was it fixed?