Aug 2026· Bioinformatics· Vol 42· 0 citations· 33 references
Medicine
TL;DR
Among class imbalance strategies tested, Tail-Aware Focal Loss outperforms standard cross-entropy by 3.1% on Macro F1 and 7.4% on Recall, striking a better balance between class performance than either weighted cross-entropy or weighted random sampling.
Abstract
Abstract Motivation Polypharmacy is increasingly common in clinical practice, and the sheer number of possible drug combinations makes manual interaction screening impractical. Early computational approaches relied on chemical similarity metrics and rule-based systems, while subsequent machine learning and deep learning methods improved predictive power but continued to treat drugs as isolated entities, missing the broader biological context that governs interaction behaviour. Graph Neural Network (GNN) based methods address this by modeling drugs alongside proteins, diseases, and side effects in a shared relational graph, but tend to fall short on sparsely represented long-tail interaction classes due to the severe class imbalance that characterizes real-world biomedical interaction data. Results We construct a large-scale heterogeneous biomedical knowledge graph—26 408 nodes across five entity types and 1 679 387 edges across six relation types—and benchmark MLP, GCN, HGT, and RGCN for 105- class DDI prediction. RGCN achieves the strongest overall performance (Macro F1: 0.694, Recall: 0.720), with relation-specific weight matrices proving the critical factor in heterogeneous DDI modelling. Among class imbalance strategies tested, Tail-Aware Focal Loss outperforms standard cross-entropy by 3.1% on Macro F1 and 7.4% on Recall, striking a better balance between class performance than either weighted cross-entropy or weighted random sampling.
SGTL-DDA is proposed, a novel graph transformer framework designed to incorporate structural information and domain-specific knowledge from heterogeneous biological information networks (HBINs) that successfully identifies both known therapeutics and novel repositioning candidates, supported by molecular docking results and literature evidence.
Bowei Zhao, Hui Zhao, Yu-an Huang et al.· IEEE transactions on computa...· 0 citations
Graph-based machine learning is promising for DDI prioritization and hypothesis generation but remains insufficient for independent clinical decision-making, and future studies should use standardized benchmarks, leakage-aware validation, calibrated uncertainty, reproducible pipelines, validated explanations, and external or prospective evaluation.
Md. Tuhin Reza, Md. Abdul Kader, Wissem Inoubli et al.· Pharmaceuticals· 0 citations
With the increasing use of multiple medications in clinical practice, accurate and interpretable prediction of organ-level adverse drug reactions (ADRs) induced by drug combinations is essential for drug safety assessment and precision medicine. Existing knowledge graph (KG)-based methods primarily model biomedical associations but leave direct structure-level interactions within drug pairs undercharacterized, while molecular representation methods often rely on whole-molecule or latent substructure encodings, offering limited chemically meaningful evidence for ADR risks. This study proposes MolADR, a multiscale complementary learning framework that integrates GNN-based KG learning with dual-granularity molecular cross-attention modeling to combine macro-level biomedical associations with microlevel molecular interaction cues. Under an emerging-drug setting, MolADR achieves PR-AUC scores of 81.17 ± 3.97, 84.04 ± 4.67, and 74.37 ± 9.90 on three data sets, consistently outperforming state-of-the-art baselines, with further analyses supporting its robustness and suggesting its ability to highlight chemically plausible atoms and functional groups for organ-level ADR prediction.
Yifan Qi, Q. Ren, Chen-Xu Wang et al.· Journal of Chemical Informat...· 0 citations
Background Adverse drug–drug interactions (DDIs) cause preventable hospitalizations, but exhaustive experimental screening of all drug pairs is infeasible. Many computational predictors rely on SMILES or other molecular representations, limiting their direct applicability to biologics and other non-small-molecule therapeutics. We present a structure-agnostic framework that combines semantic representations derived from Medical Subject Headings (MeSH) with graph-derived topology from a Drug–Target–Protein knowledge graph constructed from DrugBank and UniProt. We further investigate how variation in MeSH annotation depth affects predictive performance. Results Drugs are grouped according to their deepest MeSH annotation level (Low, Mid, or Deep), and performance is evaluated across the resulting interaction categories in transductive and inductive settings. The Intermediate ontology scope (Low+Mid) provides the most stable performance, while adding Deep-level terms offers limited and inconsistent benefit. Lightweight topological descriptors are integrated with MeSH features through instance-wise, dimension-specific latent-space gating, using curated reliable-negative pairs for supervision. Fusion improves mean performance over the MeSH-only baseline across all six categories in the transductive setting. Under induction, the clearest gains occur for Low–Low interactions (ΔAUROC = 0.056; ΔF1 = 0.137) and Low–Mid interactions (ΔAUROC = 0.077; ΔF1 = 0.114). Conclusions MeSH annotation depth is associated with systematic variation in DDI prediction performance that aggregate evaluation can obscure. Graph-derived topology is particularly beneficial when ontology annotations are shallow. The framework provides a common, structure-agnostic representation compatible with both small-molecule and biologic therapeutics and supports first-pass DDI prioritization for subsequent expert assessment.
Experiments demonstrate that HSAF-DDI achieves superior overall performance compared to state-of-the-art methods, indicating the critical role of fine-grained biological features in improving the DDI prediction performance.
Xiaoli Lin, Si-Yuan Zhang, Bo Li et al.· Journal of Chemical Informat...· 0 citations
Background: Graph neural networks improve computational prediction of polypharmacy side effects, but standard binary cross-entropy training allocates equal capacity to well-classified and difficult examples, potentially missing clinically significant interactions. We evaluated whether an asymmetric focal objective could improve multi-relational drug-drug interaction (DDI) prediction by emphasizing difficult positive interactions. Methods: ClinicalFocal loss was integrated into a relation-aware graph convolutional network using molecular fingerprints, physicochemical descriptors, and learned embeddings. The model was evaluated on TWOSIDES using five-fold cross-validation with identical experimental conditions (architecture, features, data partitions, hyperparameters, and random seeds) for ClinicalFocal loss and binary cross-entropy baseline. Results: ClinicalFocal loss increased accuracy from 0.699 to 0.892 (+19.3 percentage points) and F1 score from 0.700 to 0.894 (+19.4 percentage points). AUROC increased from 0.766 to 0.914, and AUCPR increased from 0.714 to 0.860. The false-negative rate decreased from 29.8% to 9.1%, while specificity increased from 69.6% to 87.5%. Overall classification error decreased from 30.1% to 10.8%, corresponding to a 64.1% relative reduction. Improvements were consistent across all five folds. Conclusions: Asymmetric focal optimization improved classification and ranking performance while achieving 90.9% recall for observed interaction triples, without modifying the underlying architecture. Loss-function design is a direct, tunable lever for improving graph-based DDI prediction.
Faranak Hatami, Mousa Moradi· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
