Limitations

It is a comparator, not an affinity predictor

It returns which of two targets binds a compound more tightly. It has no opinion on whether the compound binds either one at all, so a confident ranking of two targets it barely touches is still a confident ranking.

Most of the answer is target identity

The same forest with the ligand removed scores 0.710, against 0.750 with it. The compound contributes about four points. That is a real and reproducible contribution, and it is far less than the headline suggests on its own. Read the headline against the baselines, never alone.

A near-tie is a near-tie

Where the two measured values sit within half a log the model is right 0.573 of the time, which is close to a coin flip. It reaches 0.915 only when the true separation is beyond two logs. The strength returned with a comparison is the guide to which case you are in.

Some family pairings are much harder than others

Across pairings carrying at least 100 held-out comparisons:

Pairings with fewer than 30 held-out comparisons are not quoted at all, here or in the manifest.

Strength is a population statistic

The strength returned with a comparison is not the probability that the particular answer is right. It is a ranking signal whose reliability was measured across a large held-out set. A comparison at or above 0.80 belongs to a population that is right 0.933 of the time.

It covers the targets it was fitted on

1,879 targets across 38 families are servable, and a target outside that roster is refused rather than guessed at. Targets are named by UniProt accession because gene symbols are many-to-many against accessions and the failure is silent.

Breadth across families is not a toxicity claim

A compound that ranks highly against many families is not thereby promiscuous or unsafe. Cross-family breadth in this data partly records how many assay panels a compound went through. This is an off-target triage and repurposing tool, and that is the claim the measurements support.