Where the Human Actually Decides | AI Learning Guide

Look at the list of decisions and notice how early the first one comes. Choosing which form of the antigen to present happens before any selection, and it already constrains what kind of antibody you can get. That is the pattern across this whole page: the procedures are fixed, but the choices inside them are not. The stringency of a selection round is a good example — set it too low and weak binders flood the output, set it too high and nothing survives, and there is no formula that tells you where the line is. The same applies to reading a binding signal, where experience with the target family is what separates a real binder from a cross-reactive one. And the stopping rule is a judgment too, because affinity is not the only requirement a usable antibody has to meet.

0:00 / 0:00

Auto-play

Decisions that shape the outcome

Which target, and which form of the antigen to present to the selection system
Which antibody format and scaffold to work with
How stringent to make each selection round
Whether a measured binding signal is real, cross-reactive, or an assay artifact
When the candidate is good enough to advance, given stability and manufacturing constraints

A cycle, not a pipeline

Each round of human design ends with measurements that change what the next round should be. That feedback is the mechanism by which the process improves, and it is also the reason the process is slow: the loop is bounded by how many real experiments a team can run, and every round consumes time and material.

Expert intuition here is not a vague quality. It is accumulated familiarity with how a particular target family behaves in a particular assay, which is what allows a scientist to tell a genuine binder from a cross-reactive or artifactual signal.