Behavior of quantifiability prediction on non-promoted pairs

Determine the behavior of the quantifiability-prediction model on the larger population of compound–assay pairs that were not selected for dose–response profiling, beyond the historically promoted pairs used for model training and evaluation.

Background

The study evaluates quantifiability prediction retrospectively and conditionally on the historical promotion policy: only compound–assay pairs already selected for 11-point dose–response profiling are included. Consequently, the reported predictive performance does not establish how the model behaves across the much larger set of screened pairs that were not promoted for profiling. Determining performance on this broader population would clarify the model’s applicability to the upstream triage decision and its potential use before historical promotion criteria are applied.

References

The analysis is retrospective and conditional on the historical promotion policy: the model was trained and evaluated among pairs that had already been selected for profiling, and its behavior on the much larger population of non-promoted pairs is unknown.

— Predicting Quantifiability from Primary Screens to Prioritize Dose-Response Profiling  (2608.26538 - Lim, 27 Aug 2026) in Discussion, paragraph beginning “There are important limits to this conclusion.”