TY  - JOUR
T1  - When Is an Artificial Intelligence Model Ready to Influence Decisions in Early-Stage Pharmaceutical Discovery?
A1  - Sarah Jensen
A1  - Pieter Vos
A1  - Lars Olsen
JF  - Pharmacophore
JO  - Pharmacophore
SN  - 2229-5402
Y1  - 2024
VL  - 15
IS  - 5
DO  - 10.51847/EXrfjyPbrw
SP  - 90
EP  - 99
N2  - Artificial intelligence models increasingly contribute to target assessment, molecular screening, property prediction, candidate generation, and evidence integration in early pharmaceutical discovery. However, the ability to produce accurate retrospective predictions does not establish that a model is ready to influence a consequential scientific or portfolio decision. Existing evaluation practices frequently emphasize benchmark performance while leaving the intended decision, consequences of error, data suitability, transportability, uncertainty, experimental corroboration, and human accountability insufficiently specified. This Original Decision-Readiness Model Article addresses that unresolved problem by proposing a decision-centred approach to evaluating artificial intelligence models before their outputs are permitted to shape early-discovery choices. The proposed model treats readiness as a conditional relationship among four interdependent components: decision context, consequence and risk, evidence adequacy, and human and organizational oversight. It further distinguishes levels of permissible influence, ranging from exploratory analysis to experimentally anchored and portfolio-influencing use. Under this approach, readiness cannot be inferred from a single performance measure or transferred automatically between targets, assays, chemical spaces, laboratories, projects, or decision stages. Instead, the evidentiary burden must increase with the consequences and irreversibility of the proposed use. Validation, calibrated uncertainty, transportability analysis, failure-mode testing, and explicit decision governance operate as cross-cutting requirements. The model is a conceptual contribution rather than an empirically validated maturity standard, regulatory instrument, or deployment protocol. Its purpose is to make claims about artificial intelligence readiness more precise, contestable, and proportionate to pharmaceutical decision risk. Future work must operationalize its components, test the reproducibility of readiness assignments, and determine whether decision-centred evaluation improves prospective scientific and portfolio outcomes.
UR  - https://pharmacophorejournal.com/article/when-is-an-artificial-intelligence-model-ready-to-influence-decisions-in-early-stage-pharmaceutical-xx2srruxhuoukzb
ER  -