Hypergraph Pattern Machine: Compositional Tokenization for Higher-Order Interactions

2026-05-15

AI Summary

Existing hypergraph learning methods struggle to capture compositional semantics—such as emergent or inhibitory effects—in high-order interactions, often leading to erroneous risk assessments in applications like drug combination prediction. To address this limitation, this work proposes the HyperGraph Pattern Machine (HGPM), which shifts the hypergraph learning paradigm from message passing to explicit subset combinatorial pattern learning. HGPM introduces a novel framework that tokenizes subsets, constructs a directed acyclic graph (DAG) based on inclusion relations, and employs an inclusion-aware Transformer coupled with a masked reconstruction pretraining strategy to model combinatorial logic among subsets. Evaluated across ten hypergraph benchmarks, HGPM achieves state-of-the-art or competitive performance and successfully identifies pharmacologically similar yet inhibitory drug combinations in real-world adverse event prediction, thereby overcoming critical shortcomings of current approaches.

Abstract

Hypergraphs model higher-order relations that drive real-world decisions, from drug prescriptions to recommendations. A central structural signal in such data, beyond what pairwise relations can express, is interaction compositionality: whether a higher-order relation is compositional, emergent, or inhibitory with respect to its observed or unobserved sets. In polypharmacy, the regime decides whether a drug should be dropped, kept, or excluded: a compositional drug triple can be safely simplified, an emergent triple requires all drugs jointly, and an inhibitory triple flags a drug that disrupts an existing interaction. However, existing hypergraph learning methods, which merely propagate messages over observed hyperedges, leave this compositional signal unmodeled, allowing dangerous drug combinations to slip through and be misclassified. To this end, we propose the Hypergraph Pattern Machine (HGPM), shifting the paradigm from message passing to learning the compositional pattern of subsets. It tokenizes compositional subsets, organizes them in an inclusion DAG, and trains an inclusion-aware Transformer under masked reconstruction. On ten hypergraph benchmarks, HGPM matches or exceeds state-of-the-art methods. Notably, in a real adverse-event prediction case, HGPM correctly identifies the drug addition that inhibits the side effect among feature-identical candidates, a discrimination existing methods cannot make. The code and data are in https://github.com/KryieZhao/HGPM.git.

Problem

• hypergraph
• higher-order interactions
• compositionality
• drug combinations
• interaction patterns

Innovation

• hypergraph learning
• compositional tokenization
• higher-order interactions
• inclusion-aware Transformer
• interaction compositionality