What the model is actually predicting | AI Learning Guide

The key thing to hold onto here is how narrow the prediction really is. The model is answering a specific question: does this molecule bind this target, and do its measurable properties fall in a usable range? It is not answering whether the molecule will help a person. To make any prediction at all, the model has to freeze most of biology — one target, one molecule, one set of conditions. That freezing is what makes the math possible, and it is also what limits the answer. So when you see a high score, read it as a shortlist decision, not a medical one.

0:00 / 0:00

Auto-play

A model predicts binding to a chosen target and a set of measurable properties. It does not predict what the molecule will do in a living body.

What the model treats as fixed

To make a prediction at all, the model holds most of biology constant. It assumes a single target structure, a single molecule, and an environment matching its training conditions. Those assumptions are not flaws in the model — they are the reason it can compute anything. The consequence is that the prediction is only as good as the assumptions, and a real body violates most of them.

What a high score means

A high score means the molecule looks promising against the target under the model's assumptions. It is a filter for deciding what to test next, not a verdict on whether the molecule will work as a medicine.