NPNguyen Van Phuong

Nguyen Van Phuong

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Nguyen Van Phuong has a total of 73 co-authors. Here are the top co-authors by number of shared publications:

  • Nguyen Thu Hang

    Department of Pharmacognosy, Faculty of Pharmacognosy and Traditional Medicine, Hanoi University of Pharmacy, Hanoi, Vietnam

    9 shared publications
  • Le Nguyen Thanh

    Institute of Marine Biochemistry, Vietnam Academy of Science and Technology (VAST), 18 Hoang Quoc Viet, Cau Giay, Ha Noi, Viet Nam

    4 shared publications
  • Nguyen Thu Hang

    Department of Pharmacognosy, Faculty of Pharmacognosy and Traditional Medicine, Hanoi University of Pharmacy, Hanoi, 11000, Viet Nam

    4 shared publications
  • Hang Nguyen Thu

    K72 Hanoi University of Pharmacy 11000 Hanoi Vietnam

    3 shared publications
  • Than Thi Kieu My

    Department of Pharmacognosy, Faculty of Pharmacognosy and Traditional Medicine, Hanoi University of Pharmacy, Hanoi, Vietnam

    3 shared publications
  • Hoang Trung Tung

    Department of Pharmacognosy, Faculty of Pharmacognosy and Traditional Medicine, Hanoi University of Pharmacy, Hanoi, 11000, Viet Nam

    2 shared publications

Latest Publications

  • Data‐Driven Prediction of Ultraviolet–Visible Spectra From Molecular Structure Using Attention‐Based Graph Neural Networks

    by Nguyen Van Phuong, Ngo Huy Trung, Nguyen Trung Khoa

    Predicting UV–vis spectra from molecular structure is an important but relatively underexplored task. In this study, we developed a deep learning approach based on graph neural networks (GNNs) to directly predict UV–vis spectra from SMILES representations. Several GNN architectures were evaluated, among which Attentive Fingerprint achieved the best performance with an R 2 of 0.8630, mean absolute error of 0.0389, root mean squared error of 0.0740, cosine similarity of 0.9589, Pearson correlation coefficient of 0.9521, and distance metric of 1.5758 on the test set. Further analysis showed that the model might be able to capture meaningful structure–property relationship across diverse chemical classes. The model was also evaluated on external datasets, where it maintained good predictive performance, indicating its generalizability. These results demonstrate that attention‐based GNNs provide an effective approach for UV–vis spectrum prediction and may offer useful insights into the molecular regions contributing to model predictions, with potential applications in molecular analysis, high‐throughput screening, and compound design. To facilitate practical use, we deployed the model as an interactive web application ( https://spectra‐prediction.streamlit.app/ ), which enables direct prediction of UV–vis spectra from SMILES inputs.

  • Open Access

    Target-Specific De Novo Drug Design via Fine-Tuned Language Models and Molecular Docking, Molecular Dynamics Simulation Validation

    by Phuong Nguyen

    De novo drug design remains an expensive and low-yield process, particularly for targets with limited bioactivity data. In this study, we fine-tuned the Qwen 2.5 0.5B model to treat ligand design as a sequence-to-sequence task from protein amino acid sequences to molecular SELFIES representations. A curated and balanced dataset derived from ChEMBL was used to reduce target-level data imbalance, followed by systematic evaluation using beam search and stochastic sampling. The fine-tuned model demonstrated strong ligand generation capability across multiple targets, achieving high validity, uniqueness, and novelty. Comparative analysis revealed that beam search favored conservative exploitation of learned chemical space, whereas sampling enabled broader exploration and higher novelty. Molecular docking, molecular dynamics simulations, and MM/PBSA binding free-energy calculations for representative targets showed that generated compounds formed stable complexes with energetically favorable binding profiles comparable to known inhibitors. Notably, target-specific fine-tuning under extreme low data conditions, enabled through molecular representation augmentation, successfully guided ligand generation toward relevant structural frameworks. Our work suggests that small language models can be effectively fine-tuned for structure-aware, target-specific ligand design, even under limited bioactivity information.

  • Integrating molecular representations for machine learning-based virtual screening of Glutaminyl Cyclase inhibitors

    by Hoang Manh Ha, Hoang Trung Tung, Nguyen Viet Anh, Nguyen Thanh Cong, Nguyen Thu Hang, Nguyen Van Phuong

  • Open Access

    Anti-Inflammatory Activities of Ocotillol Isolated from Tetragonula iridipennis Propolis: A Study on In Vitro and In Silico Models

    by Nguyen Thanh Cong, Nguyen Van Phuong, Do Van Hieu, Nguyen Hoang Viet, Le Nguyen Thanh

    Background/Objectives: This study evaluated the anti-inflammatory effects of ocotillol, a compound isolated from the ethanol extract of propolis of the Tetragonula iridipennis stingless bee. Through its ability to inhibit NO production in an in vitro model, it investigated the NO inhibition mechanism using network pharmacology combined with molecular docking. Methods: The NO production inhibitory activity was determined by colorimetric assay using Griess reagent. An in silico study was performed using network pharmacology analysis, molecular docking, and molecular dynamics simulations. Results: The in vitro results demonstrated that ocotillol exhibited significant anti-inflammatory effects by effectively inhibiting NO production, with an IC50 value of 20.29 ± 2.1 µg/mL. The network pharmacology analysis revealed that ocotillol targets 14 molecular sites related to NO, with TACR1 showing the best binding affinity at −10.0 kcal/mol. Molecular dynamics simulations suggest that TACR1 is a potential target. As indicated by the stable interaction profile, further validation in complex biological membranes is warranted. Conclusions: This study also provides evidence for the correlation between in vitro and in silico models, thus laying the groundwork for in vivo evaluations to confirm the anti-inflammatory mechanism of ocotillol.

  • HPLC Profile‐Based Chemometric Discrimination of Cinnamomum cassia by Geographic Origin

    by Bui Thi Lan Phuong, Hoang Trung Tung, Do Khanh Linh, Nguyen Minh Tuan, Vu Hong Phuc, Nguyen Manh Son, Nguyen Thi Kieu Anh, Nguyen Van Phuong

    Cinnamon is a valuable commodity, used in medicine and many other fields, and its quality depends on its origin. Therefore, classifying cinnamon by its growing region is essential. This study classified 134 cinnamon samples by their geographical origin based on the content of three key compounds, coumarin, cinnamic acid, and cinnamaldehyde, using machine learning techniques, including XGBoost, support vector machine, random forest, voting, stacking, and convolutional neural network models. The results indicated that the XGBoost model achieved the highest accuracy (76.47%) among the targeted approaches. It also showed that using only the concentrations of these three was insufficient to reliably distinguish cinnamon groups. To address limitations, an untargeted approach used the full HPLC chromatogram at 280 nm. This approach significantly improved classification performance, with voting and stacking models reaching an accuracy of 94.18%. SHapley Additive exPlanations analysis further revealed the retention times corresponding to the most critical signals to model predictions. On the other hand, applying principal component analysis to reduce the data dimensionality significantly improved the performance of the established model to an impressive accuracy of 100%. These results suggested an efficient approach to discriminate cinnamon origins by integrating HPLC data with machine learning.


  • Open Access

    Chemical Constituents and α-Glucosidase Inhibitory Activities of the Leaves of Embelia parviflora—In Vitro and In Silico Studies

    by Sy Danh Thuong, Mai Thi Hoang Anh, Nguyen Van Phuong, Chu Hoang Mau, Nguyen Huu Quan, Nguyen Thanh Cong, Le Nguyen Thanh

    Phytochemical investigation of the methanol extract of Embelia parviflora Wall. Ex A. DC. leaves (Primulaceae family) led to the isolation of sixteen compounds including three sterols (1–3), one triterpene (4), four flavonoids (5–8), four megastigmanes (9–12), three phenolic compounds (13–15), and one furan derivative (16). Their chemical structures were determined based on ESI-MS and NMR spectral data. This is the first chemical study of E. parviflora. Compounds 3, 8–13, and 16 were found in the Embelia genus for the first time. Compounds 9–11, 13, and 16 represent the first isolation from the Primulaceae family. In the α-glucosidase activity assay, MeOH extract, compounds 4 and 5 strongly inhibited enzyme α-glucosidase activity. A molecular docking study revealed that compounds 4 and 5 showed different interactions with enzyme α-glucosidase.


  • Enhanced prediction of beta-secretase inhibitory compounds with mol2vec technique and machine learning algorithms

    by N.T. Hang, N.D. Duy, T.D.H. Anh, L.T.N. Mai, N.T.B. Loan, N.T. Cong, N.V. Phuong

    A comprehensive computational strategy that combined QSAR modelling, molecular docking, and ADMET analysis was used to discover potential inhibitors for β-secretase 1 (BACE-1). A dataset of 1,138 compounds with established BACE-1 inhibitory activities was used to build a QSAR model using mol2vec descriptors and support vector regression. The obtained model demonstrated strong predictive performance (training set: r2 = 0.790, RMSE = 0.540, MAE = 0.362; test set: r2 = 0.705, RMSE = 0.641, MAE = 0.495), indicating its reliability in identifying potent BACE-1 inhibitors. By applying this QSAR model together with molecular docking, seven compounds (ZINC8790287, ZINC20464117, ZINC8878274, ZINC96116481, ZINC217682404, ZINC217786309 and ZINC96113994) were identified as promising candidates, exhibiting predicted log IC50 values ranging from 0.361 to 1.993 and binding energies ranging from -10.8 to -10.7 kcal/mol. Further analysis using ADMET studies and molecular dynamics simulations provided further support for the potential of compound 279 (ZINC96116481) and compound 945 (ZINC96113994) as drug candidates. However, since our study is purely theoretical, further experimental validation through in vitro and in vivo studies is essential to confirm these promising findings.


  • Open Access

    Adapting Language Models for Retrosynthesis Prediction

    by Phuong Nguyen-Van, Long Nguyen Thanh, Ha Hoang Manh, Ha Anh Pham Thi, Thanh Le Nguyen, Viet Anh Nguyen

    Retrosynthesis, a commonly used technique used in organic synthesis, involves deconstructing complex molecules into simpler precursors, thus enabling the efficient synthesis of target compounds. Recently, advancements in deep learning and large language models (LLMs) have offered a transformative approach to this domain. In this study, we presented a fine-tuning language model tailored explicitly for retrosynthesis prediction. By fine-tuning this model on extensive chemical reaction data (USPTO-50K), we successfully proposed appropriate retrosynthetic pathways for approximately 92% of cases in the test set. Through detailed case studies and step-by-step retrosynthesis predictions, our model demonstrated proficiency in identifying reaction types, centers, and reagents, albeit with occasional deviations that reflect its creative approach to reaction prediction. These findings highlight the model's potential as a valuable tool in organic synthesis planning.


  • In silico screening of Fyn kinase inhibitors using classification-based QSAR model, molecular docking, molecular dynamics and ADME study

    by Nguyen Thu Hang, Thai Doan Hoang Anh, Le Nguyen Thanh, Nguyen Viet Anh, Nguyen Van Phuong

    This study aimed to use a computational approach that combined the classification-based QSAR model, molecular docking, ADME studies, and molecular dynamics (MD) to identify potential inhibitors of Fyn kinase. First, a robust classification model was developed from a dataset of 1,078 compounds with known Fyn kinase inhibitory activity, using the XGBoost algorithm. After that, molecular docking was performed between potential compounds identified from the QSAR model and Fyn kinase to assess their binding strengths and key interactions, followed by MD simulations. ADME studies were additionally conducted to preliminarily evaluate the pharmacokinetics and drug-like characteristics of these compounds. The results showed that our obtained model exhibited good predictive performance with an accuracy of 0.95 on the test set, affirming its reliability in identifying potent Fyn kinase inhibitors. Through the application of this model in conjunction with molecular docking and ADME studies, nine compounds were identified as potential Fyn kinase inhibitors, including 208 (ZINC70708110), 728 (ZINC8792432), 734 (ZINC8792187), 736 (ZINC8792350), 738 (ZINC8792286), 739 (ZINC8792309), 817 (ZINC33901069), 852 (ZINC20759145), and 1227 (ZINC100006936). MD simulations further demonstrated that the four most promising compounds, 728, 734, 736, and 852 exhibited stable binding with Fyn kinase during the simulation process. Additionally, a web-based platform (https://fynkinase.streamlit.app/) has been developed to streamline the screening process. This platform enables users to predict the activity of their substances of interest on Fyn kinase from their SMILES, using our classification-based QSAR model and molecular docking.


  • Identification of potential FAK inhibitors using mol2vec molecular descriptor-based QSAR, molecular docking, ADMET study, and molecular dynamics simulation

    by Nguyen Thu Hang, Than Thi Kieu My, Le Thi Van Anh, Phan Thi Van Anh, Thai Doan Hoang Anh, Nguyen Van Phuong

    This study aims to identify potential focal adhesion kinase (FAK) inhibitors through an integrated computational approach, combining mol2vec descriptor-based QSAR, molecular docking, ADMET study, and molecular dynamics simulation. A dataset of 437 compounds with known FAK inhibitory activities was used to develop QSAR models using machine learning algorithms combined with mol2vec descriptors. Subsequently, the most promising compounds were subjected to molecular docking against FAK to evaluate their binding affinities and key interactions. ADMET study and molecular dynamics simulation were also employed to investigate the pharmacokinetic, drug-like properties, and the stability of the protein–ligand complexes. The results showed that the mol2vec descriptor-based QSAR model established by support vector regression demonstrated good predictive performance (R2 = 0.813, RMSE = 0.453, MAE = 0.263 in case of training set, and R2 = 0.729, RMSE = 0.635, MAE = 0.477 in case of test set), indicating their reliability in identifying potent FAK inhibitors. Using this QSAR model and molecular docking, compound 21 (ZINC000004523722) was identified as the most potential compound, with predicted logIC50 value and binding energy of 2.59 and − 9.3 kcal/mol, respectively. The results of molecular dynamics simulation and ADMET study also further suggested its potential as a promising drug candidate. However, because our research was merely theoretical, additional in vitro and in vivo studies are required for the verification of these results.