Zhifeng Gao
Zhifeng Gao has a total of 135 co-authors. Here are the top co-authors by number of shared publications:
- Guolin Ke
DP Technology, Beijing, China
12 shared publications - Hongshuai Wang
Dp technology
8 shared publications Linfeng Zhang
DP Technology, Beijing, China
6 shared publications- Linfeng Zhang
DP Technology, Beijing, China
6 shared publications - Zheng Cheng
AI for Science Institute, Beijing, China
6 shared publications Hongshuai Wang
DP Technology, Beijing
5 shared publications
Latest Publications
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 6586-6596
Residual Orthogonal Projection for Multi-Modal Molecular Retrieval
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 6586-6596
Open AccessResidual Orthogonal Projection for Multi-Modal Molecular Retrieval
- Open Access
- 8 June 2026
- Published by Springer Nature in Research Square
NMRPeak: a ready-to-use intelligent system for molecular structure elucidation enabled by synergistic cross-modal learning
One-dimensional nuclear magnetic resonance (NMR) spectroscopy is essential for molecular structure elucidation in organic synthesis, drug discovery, natural product characterization, and metabolomics, yet its interpretation remains heavily dependent on expert knowledge and difficult to scale. Although machine learning has been applied to NMR spectrum prediction, library retrieval, and structure generation, these tasks have largely evolved in isolation using simulated data and incompatible spectral representations, limiting their utility in real experimental settings. Here we present NMRPeak, a unified cross-modal learning system that integrates these three tasks through an experimentally grounded design. We curate approximately 1.8 million experimental and simulated spectra to construct the largest benchmark for NMR-based structure elucidation. Systematic benchmarking reveals that existing models trained predominantly on simulated spectra suffer severe performance degradation on experimental data, with top-1 structure elucidation accuracy falling below 20\%, underscoring a critical gap between current simulation paradigms and practical deployment. To bridge this gap, we introduce NMRPeak, a unified system built on a molecule-to-spectrum paradigm with synergistic coupling of prediction, retrieval, and generation modules. The framework is underpinned by a chemically-aware adaptive tokenizer that dynamically balances discretization granularity to preserve spectral semantics while controlling vocabulary size, and a peak-aware similarity metric that enables direct comparison across spectra without requiring peak assignments. NMRPeak overcomes the longstanding simulation-to-experiment gap in spectrum prediction while achieving over 95\% top-1 accuracy in molecular retrieval and approximately 75\% top-1 accuracy in stereochemistry-aware \textit{de novo} structure generation, substantially outperforming prior methods under experimental conditions. These capabilities establish a foundation for automated and high-throughput molecular structure elucidation in organic synthesis, drug discovery, and chemical biology, positioning NMRPeak as a robust and ready-to-use platform for AI-assisted NMR analysis.
- 8 June 2026
- Published by Springer Nature in Research Square
Open AccessNMRPeak: a ready-to-use intelligent system for molecular structure elucidation enabled by synergistic cross-modal learning
One-dimensional nuclear magnetic resonance (NMR) spectroscopy is essential for molecular structure elucidation in organic synthesis, drug discovery, natural product characterization, and metabolomics, yet its interpretation remains heavily dependent on expert knowledge and difficult to scale. Although machine learning has been applied to NMR spectrum prediction, library retrieval, and structure generation, these tasks have largely evolved in isolation using simulated data and incompatible spectral representations, limiting their utility in real experimental settings. Here we present NMRPeak, a unified cross-modal learning system that integrates these three tasks through an experimentally grounded design. We curate approximately 1.8 million experimental and simulated spectra to construct the largest benchmark for NMR-based structure elucidation. Systematic benchmarking reveals that existing models trained predominantly on simulated spectra suffer severe performance degradation on experimental data, with top-1 structure elucidation accuracy falling below 20\%, underscoring a critical gap between current simulation paradigms and practical deployment. To bridge this gap, we introduce NMRPeak, a unified system built on a molecule-to-spectrum paradigm with synergistic coupling of prediction, retrieval, and generation modules. The framework is underpinned by a chemically-aware adaptive tokenizer that dynamically balances discretization granularity to preserve spectral semantics while controlling vocabulary size, and a peak-aware similarity metric that enables direct comparison across spectra without requiring peak assignments. NMRPeak overcomes the longstanding simulation-to-experiment gap in spectrum prediction while achieving over 95\% top-1 accuracy in molecular retrieval and approximately 75\% top-1 accuracy in stereochemistry-aware \textit{de novo} structure generation, substantially outperforming prior methods under experimental conditions. These capabilities establish a foundation for automated and high-throughput molecular structure elucidation in organic synthesis, drug discovery, and chemical biology, positioning NMRPeak as a robust and ready-to-use platform for AI-assisted NMR analysis.
- Open Access
- 21 May 2026
- Published by American Chemical Society (ACS) in JACS Au
- Vol. 6  (6) , 3138-3152
Fast and Flexible 3D Molecular Design Framework for Novel Organic Optoelectronic Materials
Organic optoelectronic materials (OOMs) are pivotal for advancing technologies such as organic photovoltaics and light-emitting diodes. Traditional methods for discovering new OOMs are inefficient and limited by chemical space exploration. We introduce O 2 -GEN, a novel framework leveraging a 3D pretraining backbone trained on a diverse data set of over ten million molecules, enabling comprehensive exploration of chemical space. O 2 -GEN is effective at generating novel fused-ring systems and conjugated fragment assemblies. It significantly outperforms existing models in these specific tasks in speed and chemical structural validity, particularly for larger molecules. The framework supports both global and local generation modes, allowing for the design of new molecules or modifications of existing molecules. Additionally, O 2 -GEN integrates a property selector fine-tuned with density functional theory data, enabling precise multiproperty screening. Overall, our O 2 -GEN allows the construction of data sets with multiproperty-biased distributions tailored to specific application scenarios, facilitating the discovery of novel optoelectronic materials.
- 21 May 2026
- Published by American Chemical Society (ACS) in JACS Au
- Vol. 6  (6) , 3138-3152
Open AccessFast and Flexible 3D Molecular Design Framework for Novel Organic Optoelectronic Materials
Organic optoelectronic materials (OOMs) are pivotal for advancing technologies such as organic photovoltaics and light-emitting diodes. Traditional methods for discovering new OOMs are inefficient and limited by chemical space exploration. We introduce O 2 -GEN, a novel framework leveraging a 3D pretraining backbone trained on a diverse data set of over ten million molecules, enabling comprehensive exploration of chemical space. O 2 -GEN is effective at generating novel fused-ring systems and conjugated fragment assemblies. It significantly outperforms existing models in these specific tasks in speed and chemical structural validity, particularly for larger molecules. The framework supports both global and local generation modes, allowing for the design of new molecules or modifications of existing molecules. Additionally, O 2 -GEN integrates a property selector fine-tuned with density functional theory data, enabling precise multiproperty screening. Overall, our O 2 -GEN allows the construction of data sets with multiproperty-biased distributions tailored to specific application scenarios, facilitating the discovery of novel optoelectronic materials.
- Top Cited PapersOpen Access
- 2 April 2026
- Published by Springer Nature in Nature Communications
- Vol. 17  (1)
NMR-Solver: automated structure elucidation via large-scale spectral matching and physics-guided fragment optimization
Nuclear Magnetic Resonance (NMR) spectroscopy is one of the most powerful and widely used tools for molecular structure elucidation in organic chemistry. However, the interpretation of NMR spectra to determine unknown molecular structures remains a labor-intensive and expertise-dependent process, particularly for complex or novel compounds. Although recent methods have been proposed for molecular structure elucidation, they often underperform in real-world applications due to inherent algorithmic limitations and limited high-quality data. Here, we present NMR-Solver, a practical and interpretable framework for the automated determination of small organic molecule structures from 1 H and 13 C NMR spectra. Our method introduces an automated framework for molecular structure elucidation, integrating large-scale spectral matching with physics-guided molecular optimization that exploits atomic-level structure–spectrum relationships in NMR. We evaluate NMR-Solver on simulated benchmarks, curated experimental data from the literature, and real-world experiments, demonstrating its strong generalization, robustness, and practical utility in real-life scenarios. By integrating computational NMR analysis, deep learning, and interpretable chemical reasoning into a unified system, it facilitates scalable, automated, and chemically meaningful molecular structure elucidation, establishing a generalizable paradigm for solving inverse problems in molecular science.
- 2 April 2026
- Published by Springer Nature in Nature Communications
- Vol. 17  (1)
Top Cited PapersOpen AccessNMR-Solver: automated structure elucidation via large-scale spectral matching and physics-guided fragment optimization
Nuclear Magnetic Resonance (NMR) spectroscopy is one of the most powerful and widely used tools for molecular structure elucidation in organic chemistry. However, the interpretation of NMR spectra to determine unknown molecular structures remains a labor-intensive and expertise-dependent process, particularly for complex or novel compounds. Although recent methods have been proposed for molecular structure elucidation, they often underperform in real-world applications due to inherent algorithmic limitations and limited high-quality data. Here, we present NMR-Solver, a practical and interpretable framework for the automated determination of small organic molecule structures from 1 H and 13 C NMR spectra. Our method introduces an automated framework for molecular structure elucidation, integrating large-scale spectral matching with physics-guided molecular optimization that exploits atomic-level structure–spectrum relationships in NMR. We evaluate NMR-Solver on simulated benchmarks, curated experimental data from the literature, and real-world experiments, demonstrating its strong generalization, robustness, and practical utility in real-life scenarios. By integrating computational NMR analysis, deep learning, and interpretable chemical reasoning into a unified system, it facilitates scalable, automated, and chemically meaningful molecular structure elucidation, establishing a generalizable paradigm for solving inverse problems in molecular science.
- 19 February 2026
- Published by Elsevier in Pattern Recognition
- Vol. 177
VitaMol: Lightweight molecular modality expansion in discrete space for multi-task transfer learning
- 19 February 2026
- Published by Elsevier in Pattern Recognition
- Vol. 177
VitaMol: Lightweight molecular modality expansion in discrete space for multi-task transfer learning
- Open Access
ReGuLaR: Variational Latent Reasoning Guided by Rendered Chain-of-Thought
While Chain-of-Thought (CoT) significantly enhances the performance of Large Language Models (LLMs), explicit reasoning chains introduce substantial computational redundancy. Recent latent reasoning methods attempt to mitigate this by compressing reasoning processes into latent space, but often suffer from severe performance degradation due to the lack of appropriate compression guidance. In this study, we propose Rendered CoT-Guided variational Latent Reasoning (ReGuLaR), a simple yet novel latent learning paradigm resolving this issue. Fundamentally, we formulate latent reasoning within the Variational Auto-Encoding (VAE) framework, sampling the current latent reasoning state from the posterior distribution conditioned on previous ones. Specifically, when learning this variational latent reasoning model, we render explicit reasoning chains as images, from which we extract dense visual-semantic representations to regularize the posterior distribution, thereby achieving efficient compression with minimal information loss. Extensive experiments demonstrate that ReGuLaR significantly outperforms existing latent reasoning methods across both computational efficiency and reasoning effectiveness, and even surpasses CoT through multi-modal reasoning, providing a new and insightful solution to latent reasoning. Code: https://github.com/FanmengWang/ReGuLaR.
Open AccessReGuLaR: Variational Latent Reasoning Guided by Rendered Chain-of-Thought
While Chain-of-Thought (CoT) significantly enhances the performance of Large Language Models (LLMs), explicit reasoning chains introduce substantial computational redundancy. Recent latent reasoning methods attempt to mitigate this by compressing reasoning processes into latent space, but often suffer from severe performance degradation due to the lack of appropriate compression guidance. In this study, we propose Rendered CoT-Guided variational Latent Reasoning (ReGuLaR), a simple yet novel latent learning paradigm resolving this issue. Fundamentally, we formulate latent reasoning within the Variational Auto-Encoding (VAE) framework, sampling the current latent reasoning state from the posterior distribution conditioned on previous ones. Specifically, when learning this variational latent reasoning model, we render explicit reasoning chains as images, from which we extract dense visual-semantic representations to regularize the posterior distribution, thereby achieving efficient compression with minimal information loss. Extensive experiments demonstrate that ReGuLaR significantly outperforms existing latent reasoning methods across both computational efficiency and reasoning effectiveness, and even surpasses CoT through multi-modal reasoning, providing a new and insightful solution to latent reasoning. Code: https://github.com/FanmengWang/ReGuLaR.
- Open Access
- 8 December 2025
- Published by Springer Nature in Research Square
Unifying Polymer Modeling and Design via a Conformation-Centric Generative Foundation Model
Polymers, macromolecules formed from covalently bonded monomers, underpin countless technologies and are indispensable to modern life. While deep learning is advancing polymer science, existing methods typically represent the whole polymer solely through monomer-level descriptors, overlooking the global structural information inherent in polymer conformations, which ultimately limits their practical performance. Moreover, this field still lacks a universal foundation model that can effectively support diverse downstream tasks, thereby severely constraining progress. To address these challenges, we introduce PolyConFM, the first polymer foundation model that unifies polymer modeling and design through conformation-centric generative pretraining. Recognizing that each polymer conformation can be decomposed into a sequence of local conformations (i.e., those of its repeating units), we pretrain PolyConFM under the conditional generation paradigm, reconstructing these local conformations via masked autoregressive (MAR) modeling and further generating their orientation transformations to recover the corresponding polymer conformation. Besides, we construct the first high-quality polymer conformation dataset via molecular dynamics simulations to mitigate data sparsity, thereby enabling conformation-centric pretraining. Experiments demonstrate that PolyConFM consistently outperforms representative task-specific methods across diverse downstream tasks, equipping polymer science with a universal and powerful tool.
- 8 December 2025
- Published by Springer Nature in Research Square
Open AccessUnifying Polymer Modeling and Design via a Conformation-Centric Generative Foundation Model
Polymers, macromolecules formed from covalently bonded monomers, underpin countless technologies and are indispensable to modern life. While deep learning is advancing polymer science, existing methods typically represent the whole polymer solely through monomer-level descriptors, overlooking the global structural information inherent in polymer conformations, which ultimately limits their practical performance. Moreover, this field still lacks a universal foundation model that can effectively support diverse downstream tasks, thereby severely constraining progress. To address these challenges, we introduce PolyConFM, the first polymer foundation model that unifies polymer modeling and design through conformation-centric generative pretraining. Recognizing that each polymer conformation can be decomposed into a sequence of local conformations (i.e., those of its repeating units), we pretrain PolyConFM under the conditional generation paradigm, reconstructing these local conformations via masked autoregressive (MAR) modeling and further generating their orientation transformations to recover the corresponding polymer conformation. Besides, we construct the first high-quality polymer conformation dataset via molecular dynamics simulations to mitigate data sparsity, thereby enabling conformation-centric pretraining. Experiments demonstrate that PolyConFM consistently outperforms representative task-specific methods across diverse downstream tasks, equipping polymer science with a universal and powerful tool.
- Open Access
- 1 December 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Towards Generalizable Data-Driven Pharmacokinetics with Interpretable Neural ODEs
Accurate modeling of drug concentration--time (C--t) profiles is central to pharmacokinetics (PK) and plays a critical role in both early-stage compound selection and late-stage individualized dosing. Traditional PK model offer mechanistic interpretability but often rely on rigid assumptions, extensive parameterization, limiting their scalability across structurally diverse compounds and heterogeneous patient populations. In this work, we propose Uni-PK, a unified neural framework that combines molecular representations with neural ordinary differential equations (NODEs) embedded within a mechanistically grounded PK structure. Rather than solely predicting PK parameters, Uni-PK directly models the dynamic trajectory of drug concentrations from molecular and individual inputs, enabling end-to-end learning under data-scarce and noisy conditions. To account for inter-individual variability, we incorporate auxiliary covariates---such as species and dosing regimen---via a flexible context encoder, supporting personalized preclinical and clinical settings. Evaluated on rat and human datasets spanning multiple administration routes and physiological states, Uni-PK showed robust performance while remaining consistent with established pharmacokinetic principles. By integrating chemical structure and individual-specific information within a dynamic modeling framework, Uni-PK offers a scalable, interpretable, and animal-sparing solution for next-generation PK modeling and precision therapeutics.
- 1 December 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessTowards Generalizable Data-Driven Pharmacokinetics with Interpretable Neural ODEs
Accurate modeling of drug concentration--time (C--t) profiles is central to pharmacokinetics (PK) and plays a critical role in both early-stage compound selection and late-stage individualized dosing. Traditional PK model offer mechanistic interpretability but often rely on rigid assumptions, extensive parameterization, limiting their scalability across structurally diverse compounds and heterogeneous patient populations. In this work, we propose Uni-PK, a unified neural framework that combines molecular representations with neural ordinary differential equations (NODEs) embedded within a mechanistically grounded PK structure. Rather than solely predicting PK parameters, Uni-PK directly models the dynamic trajectory of drug concentrations from molecular and individual inputs, enabling end-to-end learning under data-scarce and noisy conditions. To account for inter-individual variability, we incorporate auxiliary covariates---such as species and dosing regimen---via a flexible context encoder, supporting personalized preclinical and clinical settings. Evaluated on rat and human datasets spanning multiple administration routes and physiological states, Uni-PK showed robust performance while remaining consistent with established pharmacokinetic principles. By integrating chemical structure and individual-specific information within a dynamic modeling framework, Uni-PK offers a scalable, interpretable, and animal-sparing solution for next-generation PK modeling and precision therapeutics.
- Open Access
- 16 October 2025
- Published by Springer Nature in npj Computational Materials
- Vol. 11  (1) , 1-10
Virtual characterization via knowledge-enhanced representation learning: from organic conjugated molecules to devices
- 16 October 2025
- Published by Springer Nature in npj Computational Materials
- Vol. 11  (1) , 1-10
Open AccessVirtual characterization via knowledge-enhanced representation learning: from organic conjugated molecules to devices
- Open Access
NMR-Solver: Automated Structure Elucidation via Large-Scale Spectral Matching and Physics-Guided Fragment Optimization
Nuclear Magnetic Resonance (NMR) spectroscopy is one of the most powerful and widely used tools for molecular structure elucidation in organic chemistry. However, the interpretation of NMR spectra to determine unknown molecular structures remains a labor-intensive and expertise-dependent process, particularly for complex or novel compounds. Although recent methods have been proposed for molecular structure elucidation, they often underperform in real-world applications due to inherent algorithmic limitations and limited high-quality data. Here, we present NMR-Solver, a practical and interpretable framework for the automated determination of small organic molecule structures from $^1$H and $^{13}$C NMR spectra. Our method introduces an automated framework for molecular structure elucidation, integrating large-scale spectral matching with physics-guided fragment-based optimization that exploits atomic-level structure-spectrum relationships in NMR. We evaluate NMR-Solver on simulated benchmarks, curated experimental data from the literature, and real-world experiments, demonstrating its strong generalization, robustness, and practical utility in challenging, real-life scenarios. NMR-Solver unifies computational NMR analysis, deep learning, and interpretable chemical reasoning into a coherent system. By incorporating the physical principles of NMR into molecular optimization, it enables scalable, automated, and chemically meaningful molecular identification, establishing a generalizable paradigm for solving inverse problems in molecular science.
Open AccessNMR-Solver: Automated Structure Elucidation via Large-Scale Spectral Matching and Physics-Guided Fragment Optimization
Nuclear Magnetic Resonance (NMR) spectroscopy is one of the most powerful and widely used tools for molecular structure elucidation in organic chemistry. However, the interpretation of NMR spectra to determine unknown molecular structures remains a labor-intensive and expertise-dependent process, particularly for complex or novel compounds. Although recent methods have been proposed for molecular structure elucidation, they often underperform in real-world applications due to inherent algorithmic limitations and limited high-quality data. Here, we present NMR-Solver, a practical and interpretable framework for the automated determination of small organic molecule structures from $^1$H and $^{13}$C NMR spectra. Our method introduces an automated framework for molecular structure elucidation, integrating large-scale spectral matching with physics-guided fragment-based optimization that exploits atomic-level structure-spectrum relationships in NMR. We evaluate NMR-Solver on simulated benchmarks, curated experimental data from the literature, and real-world experiments, demonstrating its strong generalization, robustness, and practical utility in challenging, real-life scenarios. NMR-Solver unifies computational NMR analysis, deep learning, and interpretable chemical reasoning into a coherent system. By incorporating the physical principles of NMR into molecular optimization, it enables scalable, automated, and chemically meaningful molecular identification, establishing a generalizable paradigm for solving inverse problems in molecular science.