Spectral Confidence Should Travel with Every Computationally Annotated Natural Product from Raw Signal Interpretation to Structure Assignment and Pharmacological Modeling

Download PDF

Computational mass spectrometry increasingly converts complex natural-product signals into molecular formulas, chemical classes, candidate structures, and named database records that can subsequently enter pharmacological modeling. Yet the evidential status of a molecular assignment is often compressed during this transition, so that a structure inferred from an imperfect tandem spectrum may eventually be represented downstream in the same structural format as a compound established with substantially stronger analytical evidence. This article develops an original evidence framework for treating spectral confidence as a persistent property of the molecular record rather than a temporary annotation-stage score. The analysis separates observed spectral evidence from formula inference, substructure or class inference, candidate generation, candidate ranking, and full-structure assignment, and it argues that confidence should remain linked to the analytical state, computational operation, candidate space, and provenance from which a structure was produced. The proposed framework does not assume that existing confidence scores are interchangeable or that uncertain annotations should be discarded. Instead, it conceptualizes confidence as typed, revisable evidence that can accompany computationally annotated molecules into databases and downstream modeling while remaining distinguishable from predictive uncertainty generated by pharmacological models themselves. This approach could support more defensible reuse, targeted reanalysis, and task-dependent handling of tentative structures. Its principal limitation is that no universal confidence scale, weighting function, or downstream performance benefit has yet been established across chemical classes, instruments, annotation systems, or pharmacological modeling tasks.