An antibody must reach its target

A binding measurement is an incomplete travel itinerary. It can make a target look inviting without telling a reader what happens between administration and the moment the drug encounters that target. For this pharmacology story, the question is deliberately practical: what must be specified before a claim about target engagement becomes meaningful?

Rules learned from simulated encounters

In February 2025, Patidar and colleagues coupled a minimal physiologically based pharmacokinetic model with decision-tree learning. They generated virtual antibody–target pairs and classified simulated occupancy. Dose schedule, antibody charge, target form and site of action changed the resulting rules. A greater-than-90% occupancy threshold was a chosen classification criterion.¹

These were simulated candidates, not a new trial showing that an antibody benefits patients. The study's framework concerns early assessment of target pharmacology.¹

Translate the endpoint before the headline

Our interpretation is that an occupancy percentage needs a location, a time and a reason for caring about it. Consider an imagined report declaring that almost all available targets are occupied. We would first ask whether the observation concerns the place where the intended effect should happen. Next, we would ask whether it describes a fleeting moment or the interval that matters for the proposed biological task.

The third question is harder: why should that extent of occupation produce the desired effect? We would want an explicit bridge between the binding endpoint and the functional endpoint. A model can make the first endpoint easier to explore without supplying evidence for the second. Keeping that gap visible makes the model more useful, because it tells the experimental team which link remains to be established.

An interpretable rule still has assumptions

Decision rules are attractive editorial material. They can be read, criticised and compared with a new observation. But a readable rule should not become a portable instruction merely because its branches are easy to draw. We would attach the conditions of the simulated world to every branch and resist presenting a threshold independently of those conditions.

For example, a team might use a rule to select two contrasting hypotheses for a follow-up experiment. We would ask that team to say in advance which result would count against the rule. A test designed only to find a confirming example would tell readers less than a comparison that could reveal its failure.

The next useful measurement

Our preferred development record would put the predicted encounter beside a measurement of engagement, then place both beside the downstream response. It would retain disagreements rather than compressing them into one overall success label. If the predicted encounter occurs but the intended response does not, that is a scientifically informative outcome worth reporting.

This study earns attention as a way to organise questions about antibody pharmacology. The standard we would apply to its successors is straightforward: show what was learned about the path to the target, and identify what still has to be learned after the drug gets there.