Validation

The endpoint

Tamoxifen is activated by CYP2D6 into endoxifen, the metabolite that does the work. A variant that lowers CYP2D6 activity lowers endoxifen formation. The test is whether the model, given only the variant, predicts that shift on a set of variants it has never seen during training.

The firewall

Anything involving tamoxifen or endoxifen is kept out of the training data entirely. If it leaked in, the test would be circular and the number would mean nothing. The held-out set lives behind a hard separation from the corpus, checked before the model is frozen.

Pre-registration

The analysis is written down and frozen before the patient data arrives: the model, the corpus, the primary comparison, and the success criterion, with a hash on the frozen protocol. The primary question is whether a continuous activity prediction beats the discrete per-allele score guidelines use today. Fixing all of this in advance is what stops the result from being tuned after the fact.

Honest state

It is not concluded. On the part of the range the model can represent, the signal is coherent and the confidence interval excludes zero. On the full set it is weaker, pulled down by gain-of-function alleles a multiplicative model cannot reach by construction. That is a genuine open outcome, which is exactly why the protocol is frozen rather than reported early. CYP2D6 is also a hard gene on its own terms, and the current corpus is noise-limited, both of which the protocol states rather than glosses over.

The work is being run with a clinical pharmacology group rather than scored in isolation.

Where this sits

Predicting a single variant's activity shift is the near-term, bounded question, and it is where the honest signal is. Predicting per-drug substrate specificity is the harder, unsolved one. The methodology explains why the second measures null on the current features, and the negative result is characterized rather than hidden.