Feature-Augmented Drug SuperHyperGraphs: Neural Architectures for Hierarchical Pharmaceutical Knowledge Representation

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

We introduce feature-augmented Drug SuperHyperGraphs, extending the Drug SuperHyperGraphframework of Fujita et al. [1] by incorporating feature vectors at each supervertex and neural learning operators.We define SuperHyperGraph Neural Networks (SHGNN), attention mechanisms, and message-passing protocolson this structure. We present a conditional approximation theorem for SHGNN under explicit assumptions(Assumptions 1-8), noting that the theorem is conditional upon the hierarchical decomposition assumption andshould not be interpreted as a complete universal approximation theorem for arbitrary SuperHyperGraphs.We prove AGG approximation via Deep Sets theory [30]. We introduce a SuperHyperGraph Laplacian withspectral analysis under the assumption 0 ≤ w(e) ≤ 1, explicitly noting that the theorem fails if edge weightsexceed unity. We prove convergence of message passing via the Banach fixed-point theorem with a discussionof the contraction assumption’s restrictiveness. We provide a corrected computational complexity analysis withfull derivation. We propose a detailed experimental protocol for future empirical validation on DrugBank,comparing SHGNN with HGNN and modern baselines (Hyper-SAGNN, HyGNN).2020 Mathematics Subject Classification: 05C65 (Primary), 68T07, 92C50, 68T05 Keywords: Drug SuperHyperGraph; SuperHyperGraph Neural Networks; Pharmaceutical AI; HypergraphLearning; Attention Mechanisms; Message Passing