StateFlux: Physically Validated Quantum Metabolic State Networks Reveal Limited Incremental Value for Toxicity Prediction
Year: 2026
Type: preprint
Abstract: Mechanistic molecular representations are increasingly proposed as a route to improve toxicity prediction by supplementing parent structure with protonation, metabolism, and quantum-chemical reactivity. Whether those physically richer representations provide predictive information that is genuinely orthogonal to molecular structure remains uncertain. Here, StateFlux represents each drug as a network of chemical and predicted metabolic states with node-level GFN2-xTB properties and edge-resolved changes in ionization potential, electron affinity, electrophilicity, solvation, and local reactivity. On a locked DILIrank 2.0 development set (n = 376; 156 DILI-positive and 220 DILI-negative compounds), 10 × 5 scaffold validation gave AUROC/AUPRC values of 0.8249/0.7727 for ECFP, 0.8049/0.7430 for ECFP + FluxCore, and 0.8010/0.7261 for ECFP + StateFluxCore. The FluxCore AUROC decrement relative to ECFP was -0.0199 (95% paired-bootstrap CI, -0.0341 to -0.0072). Strict temporal and scaffold-novel tests likewise showed no consistent advantage, and an independent 5,367-compound genotoxicity benchmark reproduced the result (AUROC 0.7765 for ECFP versus 0.7732 for StateFluxCore). To distinguish limited predictive utility from inaccurate low-cost quantum chemistry, 24 parent-to-metabolite transformations were re-evaluated with 144 ωB97X-D/6-31+G*/SMD single-point calculations. GFN2-xTB closely tracked DFT changes in electrophilicity (Pearson r = 0.963; MAE = 0.408 eV; 79.2% sign agreement). Thus, transformation-induced quantum reactivity can be physically meaningful without providing reliable incremental information for broad toxicity classification. These results argue that mechanistic fidelity and predictive utility should be validated separately when increasingly elaborate molecular representations are introduced into chemical machine learning.
Source: ChemRxiv
Author: Aris Tsai
Institution: Stanford University
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7214075727