Four Kinds of Evidence About a Drug | Visual AI Learning Guide

Think of the prediction node in the center as the question we want answered: will this drug fail? Four streams feed into it, and each one answers a different part of that question. Structure tells us what the molecule is. Assay data tells us how it behaves in a biological setting. Trial history tells us what happened to drugs like it in real people. Real-world and literature data fill in what controlled studies never tested. Notice that the map is not four alternatives — it is four partial views of the same candidate. Where all four agree, the prediction has something solid to stand on. Where one stream is thin or missing, the center node is being fed by less than it appears to be.

0:00 / 0:00

Auto-play

A prediction about a drug is assembled from four broad kinds of evidence, each answering a different question about the same candidate.

Molecular structure is the drug's chemical identity: the atoms and the shape they form. It answers what the molecule is, and it is available very early, often before the drug has been tested in anything living.

Biological and assay data describe how the drug behaves in a biological setting — how tightly it binds to its intended target, what it does to cells in a dish, how it is broken down, and whether it disturbs unrelated biology. This is measured, not inferred, and it is where a promising molecule starts to look risky.

Historical trial outcomes record what happened when past drugs were actually given to people: whether they worked, whether they were safe, and at what stage they were abandoned. This is the only category that captures the endpoint the prediction is trying to anticipate.

Real-world and literature data add the wider context — reports of side effects after a drug reaches the market, published studies, and clinical records. It is messier than the first three, but it covers situations no controlled study was designed to test.

No single category is sufficient. Structure alone says nothing about safety; assay results alone say nothing about whether a drug will survive contact with a real patient population. A prediction is strongest where these four sources agree, and weakest where one of them is thin.