Qian Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Qian Zhang has a total of 86 co-authors. Here are the top co-authors by number of shared publications:
- Guixu Zhang
East China Normal University, Shanghai, China
11 shared publications - Zhijian Xu
State Key Laboratory of Drug Research, Drug Discovery and Design Center, Shanghai Institute of Materia Medica
11 shared publications - Weiliang Zhu
CAS Key Laboratory of Receptor Research; State Key Laboratory of Drug Research; Drug Discovery and Design Center, Shanghai Institute of Materia Medica, Chinese Academy of Sciences, Shanghai, China
7 shared publications Aimin Zhou
School of Computer Science and Technology, Shanghai Frontiers Science Center of Molecule Intelligent Syntheses, East China Normal University, Shanghai 200241, China
6 shared publicationsGuixu Zhang
School of Computer Science and Technology, East China Normal University , 3663 Zhongshan North Road, Putuo District, Shanghai 200062
5 shared publicationsJian Wu
School of Computer Science and Technology, Shanghai Frontiers Science Center of Molecule Intelligent Syntheses, East China Normal University, Shanghai, China
5 shared publications
Latest Publications
- 8 June 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 66 (12) , 7124-7137
The Systematic Study of Spatially Conserved Salt Bridges in Protein
Salt-bridge conservation has traditionally been evaluated at the primary sequence level, leaving the persistence of three-dimensional interaction sites across homologous families largely unexplored. In this study, we developed a systematic family-level structural framework to redefine salt-bridge conservation based on spatial interaction sites rather than residue identity by mapping salt bridges onto unified SCOP2-aligned domain coordinates. We classified these interactions into three categories: classically conserved (CLA), nonclassically conserved (NOCLA; i.e., charge compensation), and nonconserved. This spatial definition enabled the identification of charge-swapped interactions that are invisible to standard sequence alignment. Through a comprehensive analysis of 5,679 protein families, we demonstrated that charge compensation is a recurrent and evolutionarily preserved mode of spatial salt-bridge conservation across homologous families. NOCLA was not uniformly distributed but instead showed marked structural-context dependence, being preferentially enriched in alpha and beta proteins (a/b) and concentrated in a limited subset of folds, particularly protein kinase-like, PLP-dependent transferase-like, TIM beta/alpha-barrel, and globin-like folds. To assess functional relevance, we integrated variant-effect predictors, including AlphaMissense and ESM-1v. Our results revealed that spatially conserved salt bridges exhibited significantly higher mutational sensitivity and functional constraint than nonconserved sites (CLA > NOCLA > nonconserved). Notably, highly sensitive NOCLA positions were also the most structurally concentrated, arising predominantly from a restricted set of folds, especially protein kinase-like folds, in contrast to the broader distribution of CLA and the highly dispersed pattern of nonconserved sites. Furthermore, molecular dynamics (MD) simulations coupled with MDPath-based mutual-information network analysis demonstrated that disruption of a representative NOCLA site significantly reorganizes long-range communication pathways within conserved catalytic regions of kinase domains. These findings suggest that spatially conserved salt bridges serve not only as local electrostatic stabilizers but also as critical dynamic coupling nodes within protein structures. Together, this study provides a three-dimensional family-level paradigm for analyzing electrostatic interactions in protein evolution and offers new mechanistic insights for interpreting variant effects and guiding structure-based drug design.
- 8 June 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 66 (12) , 7124-7137
The Systematic Study of Spatially Conserved Salt Bridges in Protein
Salt-bridge conservation has traditionally been evaluated at the primary sequence level, leaving the persistence of three-dimensional interaction sites across homologous families largely unexplored. In this study, we developed a systematic family-level structural framework to redefine salt-bridge conservation based on spatial interaction sites rather than residue identity by mapping salt bridges onto unified SCOP2-aligned domain coordinates. We classified these interactions into three categories: classically conserved (CLA), nonclassically conserved (NOCLA; i.e., charge compensation), and nonconserved. This spatial definition enabled the identification of charge-swapped interactions that are invisible to standard sequence alignment. Through a comprehensive analysis of 5,679 protein families, we demonstrated that charge compensation is a recurrent and evolutionarily preserved mode of spatial salt-bridge conservation across homologous families. NOCLA was not uniformly distributed but instead showed marked structural-context dependence, being preferentially enriched in alpha and beta proteins (a/b) and concentrated in a limited subset of folds, particularly protein kinase-like, PLP-dependent transferase-like, TIM beta/alpha-barrel, and globin-like folds. To assess functional relevance, we integrated variant-effect predictors, including AlphaMissense and ESM-1v. Our results revealed that spatially conserved salt bridges exhibited significantly higher mutational sensitivity and functional constraint than nonconserved sites (CLA > NOCLA > nonconserved). Notably, highly sensitive NOCLA positions were also the most structurally concentrated, arising predominantly from a restricted set of folds, especially protein kinase-like folds, in contrast to the broader distribution of CLA and the highly dispersed pattern of nonconserved sites. Furthermore, molecular dynamics (MD) simulations coupled with MDPath-based mutual-information network analysis demonstrated that disruption of a representative NOCLA site significantly reorganizes long-range communication pathways within conserved catalytic regions of kinase domains. These findings suggest that spatially conserved salt bridges serve not only as local electrostatic stabilizers but also as critical dynamic coupling nodes within protein structures. Together, this study provides a three-dimensional family-level paradigm for analyzing electrostatic interactions in protein evolution and offers new mechanistic insights for interpreting variant effects and guiding structure-based drug design.
- 16 February 2026
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 69 (6) , 6965-6975
Insights into Protein–Ligand Noncovalent Interaction Networks: A Database Survey and Quantum Chemistry Calculation Study
Protein-ligand binding is largely determined by multiple noncovalent interactions (NCIs), which may exhibit synergistic or antagonistic effects. Here, we applied an approach combining PDB statistics with QM/MM calculations to quantify relationships among five major NCIs: halogen bonds (XB), hydrogen bonds (HB), cation-π interactions (CP), π-π interactions (PP), and salt bridges (SB). Statistical analysis showed that 69.28% of the ligands with SB acceptors interact with protein residues, followed by HB acceptors (65.21%) and XB donors (62.76%). Energy analysis revealed the strongest antagonism between the CP acceptor and SB acceptor (median ΔESyn = 1.31 kcal/mol) and the strongest synergy between the HB donor and acceptor (-0.73 kcal/mol). The effects are enhanced when their interaction distances are R2 = 0.980). These results reveal a complicated NCI network, which provides guidance for rational ligand design.
- 16 February 2026
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 69 (6) , 6965-6975
Insights into Protein–Ligand Noncovalent Interaction Networks: A Database Survey and Quantum Chemistry Calculation Study
Protein-ligand binding is largely determined by multiple noncovalent interactions (NCIs), which may exhibit synergistic or antagonistic effects. Here, we applied an approach combining PDB statistics with QM/MM calculations to quantify relationships among five major NCIs: halogen bonds (XB), hydrogen bonds (HB), cation-π interactions (CP), π-π interactions (PP), and salt bridges (SB). Statistical analysis showed that 69.28% of the ligands with SB acceptors interact with protein residues, followed by HB acceptors (65.21%) and XB donors (62.76%). Energy analysis revealed the strongest antagonism between the CP acceptor and SB acceptor (median ΔESyn = 1.31 kcal/mol) and the strongest synergy between the HB donor and acceptor (-0.73 kcal/mol). The effects are enhanced when their interaction distances are R2 = 0.980). These results reveal a complicated NCI network, which provides guidance for rational ligand design.
- Open Access
- 7 January 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (1)
PEPNet: a two-stage point cloud framework with hierarchical embedding and antigen–antibody interaction modeling for epitope prediction
Epitope prediction is a key challenge in immunology and therapeutic antibody design. Existing computational methods rely on residue-level graph representations that fail to capture fine-grained atomic-level geometric information essential for antibody–antigen recognition. Considering that protein structure files (e.g. Protein Data Bank (PDB) files) inherently contain 3D atomic coordinates, we model proteins as atomic-level point clouds to directly preserve high-resolution spatial features . Building on this representation, we propose Point cloud-based Epitope Prediction Network(PEPNet), a two-stage point cloud framework for epitope prediction. Inspired by the natural atom-to-residue hierarchy in proteins, PEPNet employs a residue-aware hierarchical embedding module to aggregate atomic features into residue-level embeddings. To capture sequential dependencies absent in unordered point clouds, we integrate rotary positional encoding. Additionally, PEPNet leverages a BERT-style pretraining strategy with data augmentation to mitigate data scarcity, and a cross-attention decoder to explicitly model antigen–antibody interactions. Experimental results show that PEPNet achieves the best overall performance (MCC = 0.401, AUC = 0.765). Even when evaluated on AlphaFold3-predicted structures, PEPNet maintains strong robustness (MCC = 0.346), still outperforming WALLE (MCC = 0.305). These results underscore PEPNet’s potential for real-world antibody–antigen analysis and design.
- 7 January 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (1)
Open AccessPEPNet: a two-stage point cloud framework with hierarchical embedding and antigen–antibody interaction modeling for epitope prediction
Epitope prediction is a key challenge in immunology and therapeutic antibody design. Existing computational methods rely on residue-level graph representations that fail to capture fine-grained atomic-level geometric information essential for antibody–antigen recognition. Considering that protein structure files (e.g. Protein Data Bank (PDB) files) inherently contain 3D atomic coordinates, we model proteins as atomic-level point clouds to directly preserve high-resolution spatial features . Building on this representation, we propose Point cloud-based Epitope Prediction Network(PEPNet), a two-stage point cloud framework for epitope prediction. Inspired by the natural atom-to-residue hierarchy in proteins, PEPNet employs a residue-aware hierarchical embedding module to aggregate atomic features into residue-level embeddings. To capture sequential dependencies absent in unordered point clouds, we integrate rotary positional encoding. Additionally, PEPNet leverages a BERT-style pretraining strategy with data augmentation to mitigate data scarcity, and a cross-attention decoder to explicitly model antigen–antibody interactions. Experimental results show that PEPNet achieves the best overall performance (MCC = 0.401, AUC = 0.765). Even when evaluated on AlphaFold3-predicted structures, PEPNet maintains strong robustness (MCC = 0.346), still outperforming WALLE (MCC = 0.305). These results underscore PEPNet’s potential for real-world antibody–antigen analysis and design.
- 3 July 2025
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 68 (14) , 14882-14894
Hydrogen Bond-Induced Binding between Organofluorine and Protein via Fluorine Atoms: A Database Survey and Quantum Chemistry Calculation Study
Fluorination is common in drug design and may significantly enhance bioactivity, although the underlying mechanism is not elucidated. We noticed a coexisting interaction pattern, viz., F···H and F···O/N/S interactions between organofluorines and protein binding pockets via fluorine atoms. Through database analysis and quantum chemistry calculations, we revealed a hydrogen bond-induced fluorine bond (HBiFB), involving a pseudopocket formed by hydrogen bond donors and halogen bond acceptors. HBiFBs show interaction energies of -1.5 to -4.0 kcal/mol and consistently enhance ligand-protein interactions by ∼1 kcal/mol. Moreover, HBiFB possesses a positively charged σ-hole, bond critical points, intermolecular electron transfer, and intramolecular electron redistribution. These findings highlight HBiFB as a potentially generalizable noncovalent force in fluorinated drug optimization.
- 3 July 2025
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 68 (14) , 14882-14894
Hydrogen Bond-Induced Binding between Organofluorine and Protein via Fluorine Atoms: A Database Survey and Quantum Chemistry Calculation Study
Fluorination is common in drug design and may significantly enhance bioactivity, although the underlying mechanism is not elucidated. We noticed a coexisting interaction pattern, viz., F···H and F···O/N/S interactions between organofluorines and protein binding pockets via fluorine atoms. Through database analysis and quantum chemistry calculations, we revealed a hydrogen bond-induced fluorine bond (HBiFB), involving a pseudopocket formed by hydrogen bond donors and halogen bond acceptors. HBiFBs show interaction energies of -1.5 to -4.0 kcal/mol and consistently enhance ligand-protein interactions by ∼1 kcal/mol. Moreover, HBiFB possesses a positively charged σ-hole, bond critical points, intermolecular electron transfer, and intramolecular electron redistribution. These findings highlight HBiFB as a potentially generalizable noncovalent force in fluorinated drug optimization.
- 7 May 2025
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 68 (10) , 10486-10496
Enhanced Bioactivity of Natural Products by Halogenation: A Database Survey and Quantum Chemistry Calculation Study
Natural products (NPs) have long been the cornerstone of drug discovery. Halogenated organic NPs are limited, while around one-fourth of approved chemical drugs are organohalogens. This suggests that the introduction of halogens into NPs may enhance their potential for transformation into drugs. In this study, we utilized a matched molecular pair (MMP) approach alongside a database survey to investigate the impact of halogenation on this transformation. The study revealed that halogenation increased the bioactivity of 70.3% of NPs, with 50.3% exhibiting at least a 2-fold enhancement. Halogen bonds (XBs) are prevalent between organohalogens and their targets. To explore whether halogenated NPs could form XBs with their targets, computational studies were performed and demonstrated that halogenated NPs or NP-derived drugs formed strong XBs with their targets, resulting in improved binding affinities. This study highlights the considerable potential of introducing halogens into NPs as a strategic approach for enhancing bioactivity and facilitating the development of drugs.
- 7 May 2025
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 68 (10) , 10486-10496
Enhanced Bioactivity of Natural Products by Halogenation: A Database Survey and Quantum Chemistry Calculation Study
Natural products (NPs) have long been the cornerstone of drug discovery. Halogenated organic NPs are limited, while around one-fourth of approved chemical drugs are organohalogens. This suggests that the introduction of halogens into NPs may enhance their potential for transformation into drugs. In this study, we utilized a matched molecular pair (MMP) approach alongside a database survey to investigate the impact of halogenation on this transformation. The study revealed that halogenation increased the bioactivity of 70.3% of NPs, with 50.3% exhibiting at least a 2-fold enhancement. Halogen bonds (XBs) are prevalent between organohalogens and their targets. To explore whether halogenated NPs could form XBs with their targets, computational studies were performed and demonstrated that halogenated NPs or NP-derived drugs formed strong XBs with their targets, resulting in improved binding affinities. This study highlights the considerable potential of introducing halogens into NPs as a strategic approach for enhancing bioactivity and facilitating the development of drugs.
- Open Access
- 19 March 2025
- Published by Elsevier in Information Processing in Agriculture
- Vol. 12 (4) , 466-477
EConv-ViT: A strongly generalized apple leaf disease classification model based on the fusion of ConvNeXt and Transformer
- 19 March 2025
- Published by Elsevier in Information Processing in Agriculture
- Vol. 12 (4) , 466-477
Open AccessEConv-ViT: A strongly generalized apple leaf disease classification model based on the fusion of ConvNeXt and Transformer
- 20 December 2024
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 68 (1) , 706-718
F-CPI: A Multimodal Deep Learning Approach for Predicting Compound Bioactivity Changes Induced by Fluorine Substitution
Fluorine (F) substitution is a common method of drug discovery and development. However, there are no accurate approaches available for predicting the bioactivity changes after F-substitution, as the effect of substitution on the interactions between compounds and proteins (CPI) remains a mystery. In this study, we constructed a data set with 111,168 pairs of fluorine-substituted and nonfluorine-substituted compounds. We developed a multimodal deep learning model (F-CPI). In comparison with traditional machine learning and popular CPI task models, the accuracy, precision, and recall of F-CPI (∼90, ∼79, and ∼45%) were higher than those of GraphDTA (∼86, ∼58, and ∼40%). The application of the F-CPI for the structural optimization of hit compounds against SARS-CoV-2 3CLpro by F-substitution achieved a more than 100-fold increase in bioactivity (IC50: 0.23 μM vs 28.19 μM). Therefore, the multimodal deep learning model F-CPI would be a veritable and effective tool in the context of drug discovery and design.
- 20 December 2024
- Published by American Chemical Society (ACS) in Journal of Medicinal Chemistry
- Vol. 68 (1) , 706-718
F-CPI: A Multimodal Deep Learning Approach for Predicting Compound Bioactivity Changes Induced by Fluorine Substitution
Fluorine (F) substitution is a common method of drug discovery and development. However, there are no accurate approaches available for predicting the bioactivity changes after F-substitution, as the effect of substitution on the interactions between compounds and proteins (CPI) remains a mystery. In this study, we constructed a data set with 111,168 pairs of fluorine-substituted and nonfluorine-substituted compounds. We developed a multimodal deep learning model (F-CPI). In comparison with traditional machine learning and popular CPI task models, the accuracy, precision, and recall of F-CPI (∼90, ∼79, and ∼45%) were higher than those of GraphDTA (∼86, ∼58, and ∼40%). The application of the F-CPI for the structural optimization of hit compounds against SARS-CoV-2 3CLpro by F-substitution achieved a more than 100-fold increase in bioactivity (IC50: 0.23 μM vs 28.19 μM). Therefore, the multimodal deep learning model F-CPI would be a veritable and effective tool in the context of drug discovery and design.
- 26 July 2024
- Published by Elsevier in Computational Biology and Chemistry
- Vol. 112, 108137
MMCL-CPI: A multi-modal compound-protein interaction prediction model incorporating contrastive learning pre-training
- 26 July 2024
- Published by Elsevier in Computational Biology and Chemistry
- Vol. 112, 108137
MMCL-CPI: A multi-modal compound-protein interaction prediction model incorporating contrastive learning pre-training
- Open Access
- 27 March 2024
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 25 (3)
D3EGFR: a webserver for deep learning-guided drug sensitivity prediction and drug response information retrieval for EGFR mutation-driven lung cancer
As key oncogenic drivers in non-small-cell lung cancer (NSCLC), various mutations in the epidermal growth factor receptor (EGFR) with variable drug sensitivities have been a major obstacle for precision medicine. To achieve clinical-level drug recommendations, a platform for clinical patient case retrieval and reliable drug sensitivity prediction is highly expected. Therefore, we built a database, D3EGFRdb, with the clinicopathologic characteristics and drug responses of 1339 patients with EGFR mutations via literature mining. On the basis of D3EGFRdb, we developed a deep learning-based prediction model, D3EGFRAI, for drug sensitivity prediction of new EGFR mutation-driven NSCLC. Model validations of D3EGFRAI showed a prediction accuracy of 0.81 and 0.85 for patients from D3EGFRdb and our hospitals, respectively. Furthermore, mutation scanning of the crucial residues inside drug-binding pockets, which may occur in the future, was performed to explore their drug sensitivity changes. D3EGFR is the first platform to achieve clinical-level drug response prediction of all approved small molecule drugs for EGFR mutation-driven lung cancer and is freely accessible at https://www.d3pharma.com/D3EGFR/index.php.
- 27 March 2024
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 25 (3)
Open AccessD3EGFR: a webserver for deep learning-guided drug sensitivity prediction and drug response information retrieval for EGFR mutation-driven lung cancer
As key oncogenic drivers in non-small-cell lung cancer (NSCLC), various mutations in the epidermal growth factor receptor (EGFR) with variable drug sensitivities have been a major obstacle for precision medicine. To achieve clinical-level drug recommendations, a platform for clinical patient case retrieval and reliable drug sensitivity prediction is highly expected. Therefore, we built a database, D3EGFRdb, with the clinicopathologic characteristics and drug responses of 1339 patients with EGFR mutations via literature mining. On the basis of D3EGFRdb, we developed a deep learning-based prediction model, D3EGFRAI, for drug sensitivity prediction of new EGFR mutation-driven NSCLC. Model validations of D3EGFRAI showed a prediction accuracy of 0.81 and 0.85 for patients from D3EGFRdb and our hospitals, respectively. Furthermore, mutation scanning of the crucial residues inside drug-binding pockets, which may occur in the future, was performed to explore their drug sensitivity changes. D3EGFR is the first platform to achieve clinical-level drug response prediction of all approved small molecule drugs for EGFR mutation-driven lung cancer and is freely accessible at https://www.d3pharma.com/D3EGFR/index.php.
- 16 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-13
LSRFormer: Efficient Transformer Supply Convolutional Neural Networks With Global Information for Aerial Image Segmentation
Both local context and global context information are essential for the semantic segmentation of aerial images. Convolutional Neural Networks (CNNs) can capture local context information well but cannot model the global dependencies. Vision transformers (ViTs) are good at extracting global information but cannot retain the spatial details well. In order to leverage the advantages of these two paradigms, we integration them in one model in this study. However, global token interaction of ViT brings high computational cost, which makes it difficult to apply to large-sized aerial images. To handle this problem, we propose a novel efficient ViT block named long-short-range transformer (LSRFormer). Instead of mainstream ViTs designed as backbones, LSRFormer is a pre-training-free and plug-and-play module to be appended after CNN stages to supplement the global information. It is composed of long-range self-attention (LR-SA), short-range self-attention (SR-SA), and multi-scale-convolutional feed-forward-network (MSC-FFN). LR-SA establishes long-range dependencies at the junction of the windows and SR-SA diffuses the long-range information from window boundary to internal. MSC-FFN can capture multi-scale information inside the ViT block. We append LSRFormer block after each CNN stage of a pure convolutional network to build a model named ConvLSR-Net. Compared with existing models which combining CNN and ViTs, our model can learn both local and global representation at all stages of the model. In particular, ConvLSR-Net achieves state-of-the-art (SOTA) results on four challenging aerial image segmentation benchmarks, including iSAID, LoveDA, ISPRS Potsdam and Vaihingen. Code has been released at https://github.com/stdcoutzrh/ConvLSR-Net.
- 16 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-13
LSRFormer: Efficient Transformer Supply Convolutional Neural Networks With Global Information for Aerial Image Segmentation
Both local context and global context information are essential for the semantic segmentation of aerial images. Convolutional Neural Networks (CNNs) can capture local context information well but cannot model the global dependencies. Vision transformers (ViTs) are good at extracting global information but cannot retain the spatial details well. In order to leverage the advantages of these two paradigms, we integration them in one model in this study. However, global token interaction of ViT brings high computational cost, which makes it difficult to apply to large-sized aerial images. To handle this problem, we propose a novel efficient ViT block named long-short-range transformer (LSRFormer). Instead of mainstream ViTs designed as backbones, LSRFormer is a pre-training-free and plug-and-play module to be appended after CNN stages to supplement the global information. It is composed of long-range self-attention (LR-SA), short-range self-attention (SR-SA), and multi-scale-convolutional feed-forward-network (MSC-FFN). LR-SA establishes long-range dependencies at the junction of the windows and SR-SA diffuses the long-range information from window boundary to internal. MSC-FFN can capture multi-scale information inside the ViT block. We append LSRFormer block after each CNN stage of a pure convolutional network to build a model named ConvLSR-Net. Compared with existing models which combining CNN and ViTs, our model can learn both local and global representation at all stages of the model. In particular, ConvLSR-Net achieves state-of-the-art (SOTA) results on four challenging aerial image segmentation benchmarks, including iSAID, LoveDA, ISPRS Potsdam and Vaihingen. Code has been released at https://github.com/stdcoutzrh/ConvLSR-Net.