Drug–drug interaction extraction based on multimodal feature fusion by Transformer and BiGRU DOI Creative Commons

Chang-Qing Yu,

Shanwen Zhang, Xuqi Wang

et al.

Frontiers in Drug Discovery, Journal Year: 2024, Volume and Issue: 4

Published: Oct. 29, 2024

Understanding drug–drug interactions (DDIs) plays a vital role in the fields of drug disease treatment, development, preventing medical error, and controlling health care-costs. Extracting potential from biomedical corpora is major complement existing DDIs. Most DDI extraction (DDIE) methods do not consider graph structure molecules, which can improve performance DDIE. Considering different advantages bi-directional gated recurrent units (BiGRU), Transformer, attention mechanisms DDIE tasks, multimodal feature fusion model combining BiGRU Transformer (BiGGT) here constructed for In BiGGT, vector embeddings corpora, molecule topology graphs, are conducted by Word2vec, Mol2vec, GCN, respectively. multi-head self-attention (MHSA) integrated into to extract local–global contextual features, important The extensive experiment results on DDIExtraction 2013 shared task dataset show that BiGGT-based method outperforms state-of-the-art approaches with precision 78.22%. BiGGT expands application deep learning field

Language: Английский

The future of pharmaceuticals: Artificial intelligence in drug discovery and development DOI Creative Commons
Chen Fu, Qi Chen

Journal of Pharmaceutical Analysis, Journal Year: 2025, Volume and Issue: unknown, P. 101248 - 101248

Published: Feb. 1, 2025

Language: Английский

Citations

3
EnGCI: enhancing GPCR-compound interaction prediction via large molecular models and KAN network DOI Creative Commons
Weihao Liu, Xiaoli Li, Bo Hang

et al.

BMC Biology, Journal Year: 2025, Volume and Issue: 23(1)

Published: May 15, 2025

Abstract Background Identifying GPCR-compound interactions (GCI) plays a significant role in drug discovery and chemogenomics. Machine learning, particularly deep has become increasingly influential this domain. Large molecular models, due to their ability capture detailed structural functional information, have shown promise enhancing the predictive accuracy of downstream tasks. Consequently, exploring performance these models GCI prediction, as well evaluating effectiveness when integrated with other learning emerged compelling research area. This paper aims investigate challenges. Results study introduces EnGCI, novel model comprising two distinct modules. The MSBM integrates graph isomorphism network (GIN) convolutional neural (CNN) extract features from GPCRs compounds, respectively. These are then processed by Kolmogorov-Arnold (KAN) for decision-making. LMMBM utilizes large-scale pre-trained compounds GPCRs, subsequently, KAN is again employed Each module leverages different sources multimodal fusion enhances overall interaction prediction. Evaluating EnGCI on rigorously curated dataset, we achieved an AUC approximately 0.89, significantly outperforming current state-of-the-art benchmark models. Conclusions complementary modules: one that learns scratch prediction task, another extracts using large After further processing integration, information enable more profound exploration understanding complex relationships between compounds. offers robust efficient framework capabilities potential contribute GPCR discovery.

Language: Английский

Citations

0
Advanced deep learning approaches enable high-throughput biological and biomedicine data analysis DOI
Leyi Wei

Methods, Journal Year: 2024, Volume and Issue: 230, P. 116 - 118

Published: Aug. 22, 2024

Language: Английский

Citations

0
Drug–drug interaction extraction based on multimodal feature fusion by Transformer and BiGRU DOI Creative Commons

Chang-Qing Yu,

Shanwen Zhang, Xuqi Wang

et al.

Frontiers in Drug Discovery, Journal Year: 2024, Volume and Issue: 4

Published: Oct. 29, 2024

Understanding drug–drug interactions (DDIs) plays a vital role in the fields of drug disease treatment, development, preventing medical error, and controlling health care-costs. Extracting potential from biomedical corpora is major complement existing DDIs. Most DDI extraction (DDIE) methods do not consider graph structure molecules, which can improve performance DDIE. Considering different advantages bi-directional gated recurrent units (BiGRU), Transformer, attention mechanisms DDIE tasks, multimodal feature fusion model combining BiGRU Transformer (BiGGT) here constructed for In BiGGT, vector embeddings corpora, molecule topology graphs, are conducted by Word2vec, Mol2vec, GCN, respectively. multi-head self-attention (MHSA) integrated into to extract local–global contextual features, important The extensive experiment results on DDIExtraction 2013 shared task dataset show that BiGGT-based method outperforms state-of-the-art approaches with precision 78.22%. BiGGT expands application deep learning field

Language: Английский

Citations

0