From Shortlist to Bench | AI Learning Guide

Everything up to this point has been prediction. The shortlist is a set of educated guesses, and the only way to find out whether a guess is right is to make the molecule and measure it. Synthesis is the first real gate: a chemist has to build the compound from starting materials, and if the route is impractical, the candidate is dropped no matter how good its score looked. Once the sample exists, two different questions get asked. Binding assays measure how tightly the molecule actually holds the target, replacing the predicted number with an observed one. Cell-based tests ask whether the molecule survives in a biological setting — whether it dissolves, whether it stays intact, whether it harms the cells. Notice that these are separate questions. A molecule can bind tightly and still fail in a cell, and that distinction matters for what the team does next.

0:00 / 0:00

Auto-play

Synthesis comes before testing

A candidate cannot be evaluated until it exists as a physical sample. Chemists build shortlisted molecules from available starting materials, and the number of practical synthetic steps matters. A molecule requiring an impractical route is set aside even if its predicted score was the best in the pool — synthesizability was already a filter, and the laboratory is where that filter becomes concrete.

What the lab measures

Binding assays

Measure how tightly the molecule actually holds the target
Replace the predicted affinity with an observed value
Reveal whether the molecule reaches and occupies the binding pocket

Cell-based tests

Check behaviour in a biological setting rather than in isolation
Test solubility and stability under realistic conditions
Flag compounds that damage cells or fail to penetrate them

A test result is rarely a clean yes or no. A candidate may bind more weakly than predicted, bind as expected but degrade quickly, or fail to enter cells at all. Each of these is a distinct measurement, and each points to a different problem.