Livoa LogoLivoa
Predicting Drug–Drug Interactions Based on Integrated Similarity and Semi-Supervised Learning (DDI-IS-SL) – Yan et al.


Year: 2022    Journal: IEEE/ACM TCBB

Methodology: Integrated similarities (chemical, biological, phenotypic) + RLS with semi-supervised learning

Dataset: TWOSIDES + PubChem, DrugBank, KEGG, SIDER, OFFSIDES

Metrics: AUC, AUPR

Pros: High accuracy, efficient with sparse data. Cons: Misses nonlinear patterns, limited biological context.

HDN-DDI: A Hierarchical Molecular Graph Neural Network for Drug–Drug Interaction Prediction – Zhang et al.


Year: 2023    Journal: BMC Bioinformatics

Methodology: Hierarchical molecular graphs + attention and co-attention

Dataset: DrugBank, ZINC, BIOSNAP

Metrics: AUC, F1, Precision, Recall

Pros: Accurate, works in cold-start, interpretable substructures. Cons: Lacks biological features, scalability issues.

SumGNN: Multi-typed Drug Interaction Prediction via Efficient Knowledge Graph Summarization – Yu et al.


Year: 2021    Journal: Bioinformatics

Methodology: KG subgraph extraction + self-attention summarization

Dataset: DrugBank, TWOSIDES, BIOSNAP

Metrics: AUC, AUPR, F1

Pros: SOTA performance, interpretable, scalable. Cons: Relies on KG quality, sensitive to noisy/incomplete data.

KnowDDI: Accurate and Interpretable Drug–Drug Interaction Prediction Enabled by Knowledge Subgraph Learning – Chen et al.


Year: 2023    Journal: arXiv

Methodology: Pair-specific KG subgraph learning + drug-flow modeling

Dataset: DrugBank, KEGG, BIOSNAP KG

Metrics: AUC, AUPR, F1

Pros: Interpretable, robust on sparse graphs. Cons: High computation cost, still KG-dependent, less molecular focus.

table

by jj

0
0 uses