Selected work

Representative Papers

sHGCN: Simplified hyperbolic graph convolutional neural networks

Jun 17, 2025

To address the high computational cost and limited accuracy of hyperbolic graph neural networks (HGNNs) in modeling graphs on hyperbolic spaces, this paper proposes a Simplified Hyperbolic Graph Convolutional Network (S-HGCN). Methodologically, it introduces the first systematic simplification of core operations in the Poincaré ball model: lightweight hyperbolic exponential and logarithmic maps are designed, and the message propagation and aggregation mechanisms are reformulated to eliminate expensive geodesic distance computations. Crucially, these simplifications preserve low-distortion hyperbolic embeddings while substantially reducing time complexity. Extensive experiments demonstrate that S-HGCN achieves an average 2.3× speedup and a 1.8% improvement in accuracy across multiple graph learning benchmarks. By reconciling efficiency with expressiveness, S-HGCN establishes a new paradigm for scalable hyperbolic graph representation learning.

0 citationsRead paper

Addressing Model Overcomplexity in Drug-Drug Interaction Prediction With Molecular Fingerprints

Mar 30, 2025

To address high model complexity, poor generalizability, and excessive computational cost in drug–drug interaction (DDI) prediction, this work proposes a lightweight neural network–based minimalist modeling paradigm. Methodologically, we systematically evaluate three molecular representations—Morgan fingerprints, GCN-based graph embeddings, and MoLFormer molecular transformer embeddings—and construct compact fully connected models under leak-proof data splitting. We further integrate Grad-CAM–style gradient analysis for interpretability validation. Key contributions include: (1) demonstrating that simple molecular representations achieve state-of-the-art performance (AUC > 0.92 on DrugBank/FDA datasets) even under stringent generalization settings; (2) the first identification—via interpretable analysis—of clinically relevant pharmacophores, including CYP inhibition and P-glycoprotein substrate motifs; and (3) challenging the “complexity bias” in DDI modeling by advocating a new paradigm prioritizing data quality and incremental model development over architectural sophistication.

0 citationsRead paper

Recent publications

Latest Papers

sHGCN: Simplified hyperbolic graph convolutional neural networks

Jun 17, 2025

To address the high computational cost and limited accuracy of hyperbolic graph neural networks (HGNNs) in modeling graphs on hyperbolic spaces, this paper proposes a Simplified Hyperbolic Graph Convolutional Network (S-HGCN). Methodologically, it introduces the first systematic simplification of core operations in the Poincaré ball model: lightweight hyperbolic exponential and logarithmic maps are designed, and the message propagation and aggregation mechanisms are reformulated to eliminate expensive geodesic distance computations. Crucially, these simplifications preserve low-distortion hyperbolic embeddings while substantially reducing time complexity. Extensive experiments demonstrate that S-HGCN achieves an average 2.3× speedup and a 1.8% improvement in accuracy across multiple graph learning benchmarks. By reconciling efficiency with expressiveness, S-HGCN establishes a new paradigm for scalable hyperbolic graph representation learning.

0 citationsRead paper

Addressing Model Overcomplexity in Drug-Drug Interaction Prediction With Molecular Fingerprints

Mar 30, 2025

To address high model complexity, poor generalizability, and excessive computational cost in drug–drug interaction (DDI) prediction, this work proposes a lightweight neural network–based minimalist modeling paradigm. Methodologically, we systematically evaluate three molecular representations—Morgan fingerprints, GCN-based graph embeddings, and MoLFormer molecular transformer embeddings—and construct compact fully connected models under leak-proof data splitting. We further integrate Grad-CAM–style gradient analysis for interpretability validation. Key contributions include: (1) demonstrating that simple molecular representations achieve state-of-the-art performance (AUC > 0.92 on DrugBank/FDA datasets) even under stringent generalization settings; (2) the first identification—via interpretable analysis—of clinically relevant pharmacophores, including CYP inhibition and P-glycoprotein substrate motifs; and (3) challenging the “complexity bias” in DDI modeling by advocating a new paradigm prioritizing data quality and incremental model development over architectural sophistication.

0 citationsRead paper