ZCZheng Cheng

Zheng Cheng

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

Zheng Cheng has a total of 87 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Residual Orthogonal Projection for Multi-Modal Molecular Retrieval

    by Gengmo Zhou, Zhen Wang, Guojiang Zhao, Zheng Cheng, Xiaohong Ji, Hengxing Cai, Zhewei Wei, Zhifeng Gao

  • Open Access

    MIM-Flow: An Integrated Forward-Inverse Framework for Navigating Sparse Pareto-Optimal Regions in Functional Conjugated Molecules

    by Yecheng Wang, Guojiang Zhao, Shangqian Chen, Yuntao Ding, Gengmo Zhou, Zhifeng Gao, Zifeng Zhao, Zheng Cheng

    Functional molecules that simultaneously satisfy multiple, often competing, objectives are critical for improving device performance, yet discovering such candidates under multi-objective constraints remains a fundamental challenge. To address this, we introduce MIM-Flow (Moiety-in-Molecule FlowNet), an integrated framework supporting both accurate forward characterization and moiety-assembly-based inverse generation. MIM-Flow integrate moiety types, site types, and assembly-pattern selections into a unified MIM action space, and utilize a dual-tower architecture to predict the representations of intermediate 1 states and local sites. The dual-tower model is pretrained on 10-million-scale conformations and single-step assembly trajectories, achieving global awareness of 3D space and site distribution. Using narrowband MR-TADF emitters as validation cases, we optimize the probability flow of the assembly process within the Sub-Trajectory Balance (SubTB) GFlowNet architecture. For de novo generation, MIM-Flow predicts emission spectra with high accuracy (Rš = 0.837 against quantum-mechanical (QM) calculations). The top-3 generated molecules satisfy all multi-objective constraints and fall within the top 5% of reported MRTADF molecules. For local generation, we incorporate a synthetic-accessibility prior by restricting MIMs to reported patented scaffolds, and successfully synthesize the top-1 candidate—an unprecedented molecule whose experimental emission spectrum closely matches the QM prediction and whose CIE coordinates fall within the top 50% of reported MR-TADF molecules. Therefore, MIM-Flow can effectively explores the sparse, distributed Pareto-optimal regions that jointly satisfy property and synthesizability criteria, drastically reducing the trial-anderror space for experimental synthesis. This generalizable framework can also be transferred to the multi-objective design of other complex functional materials.

  • Open Access

    MIM-Flow: An Integrated Forward-Inverse Framework for Navigating Sparse Pareto-Optimal Regions in Functional Conjugated Molecules

    by Yecheng Wang, Guojiang Zhao, Shangqian Chen, Yuntao Ding, Gengmo Zhou, Zhifeng Gao, Zifeng Zhao, Zheng Cheng

    Functional molecules that simultaneously satisfy multiple, often competing, objectives are critical for improving device performance, yet discovering such candidates under multi-objective constraints remains a fundamental challenge. To address this, we introduce MIM-Flow (Moiety-in-Molecule FlowNet), an integrated framework supporting both accurate forward characterization and moiety-assembly-based inverse generation. MIM-Flow integrate moiety types, site types, and assembly-pattern selections into a unified MIM action space, and utilize a dual-tower architecture to predict the representations of intermediate 1 states and local sites. The dual-tower model is pretrained on 10-million-scale conformations and single-step assembly trajectories, achieving global awareness of 3D space and site distribution. Using narrowband MR-TADF emitters as validation cases, we optimize the probability flow of the assembly process within the Sub-Trajectory Balance (SubTB) GFlowNet architecture. For de novo generation, MIM-Flow predicts emission spectra with high accuracy (Rš = 0.837 against quantum-mechanical (QM) calculations). The top-3 generated molecules satisfy all multi-objective constraints and fall within the top 5% of reported MRTADF molecules. For local generation, we incorporate a synthetic-accessibility prior by restricting MIMs to reported patented scaffolds, and successfully synthesize the top-1 candidate—an unprecedented molecule whose experimental emission spectrum closely matches the QM prediction and whose CIE coordinates fall within the top 50% of reported MR-TADF molecules. Therefore, MIM-Flow can effectively explores the sparse, distributed Pareto-optimal regions that jointly satisfy property and synthesizability criteria, drastically reducing the trial-anderror space for experimental synthesis. This generalizable framework can also be transferred to the multi-objective design of other complex functional materials.

  • Open Access

    Fast and Flexible 3D Molecular Design Framework for Novel Organic Optoelectronic Materials

    by Guojiang Zhao, Zheng Cheng, Kele Xu, Tianjiao Wan, Ming Feng, Shuqi Lu, Hongshuai Wang, Zijian Gao, Qi Ou, Liang Hu, et al.

    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.

  • A Hybrid Physics-Driven Neural Network Force Field for Liquid Electrolytes

    by Junmin Chen, Qian Gao, Yange Lin, Miaofei Huang, Zheng Cheng, Wei Feng, Jianxing Huang, Bo Wang, Kuang Yu

    Electrolyte design plays an important role in the development of lithium-ion batteries and sodium-ion batteries. Battery electrolytes feature a large design space composed of different solvents, additives, and salts, which is difficult to explore experimentally. High-fidelity molecular simulation can accurately predict the bulk properties of electrolytes by employing accurate potential energy surfaces, thus guiding the molecule and formula engineering. At present, the overly simplified classic force fields rely heavily on experimental data for fine-tuning, thus its predictive power on microscopic level is under question. In contrast, the newly emerged machine learning interatomic potential (MLIP) can accurately reproduce the ab initio data, demonstrating excellent fitting ability. However, it is still haunted by problems such as low transferability, insufficient stability in the prediction of bulk properties, and poor training cost scaling. Therefore, it cannot yet be used as a robust and universal tool for the exploration of electrolyte design space. In this work, we introduce a highly scalable and fully bottom-up force field construction strategy called PhyNEO-Electrolyte. It adopts a hybrid physics-driven and data-driven method that relies only on monomer and dimer EDA (energy decomposition analysis) data. With a careful separation of long/short-range and nonbonding/bonding interactions, we rigorously restore the long-range asymptotic behavior, which is critical in the description of electrolyte systems. Through this approach, we significantly improve the data efficiency of MLIP training, allowing us to achieve much larger chemical space coverage using much less data while retaining reliable quantitative prediction power in bulk phase calculations. PhyNEO-Electrolyte thus serves as an important tool for future electrolyte optimization.


  • Open Access

    Virtual characterization via knowledge-enhanced representation learning: from organic conjugated molecules to devices

    by Guojiang Zhao, Qi Ou, Zifeng Zhao, Shangqian Chen, Haitao Lin, Xiaohong Ji, Zhen Wang, Hongshuai Wang, Hengxing Cai, Lirong Wu, et al.


  • Open Access

    FAST AND FLEXIBLE 3D MOLECULE DESIGN FRAMEWORK FOR NOVEL ORGANIC OPTOELECTRONIC MATERIALS

    by Guojiang Zhao, Zheng Cheng, Kele Xu, Tianjiao Wan, Ming Feng, Shuqi Lu, Hongshuai Wang, Zijian Gao, Qi Ou, Liang Hu, et al.

    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 O2-GEN, a novel framework leveraging a 3D pretraining backbone trained on a diverse dataset of over ten million molecules, enabling comprehensive exploration of chemical space. O2-GEN is effective at generating novel fused-ring systems and conjugated fragment assemblies. It significantly outperforms existing models 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, O2-GEN integrates a property selector fine-tuned with density functional theory data, enabling precise multi-property screening. Overall, our O2-GEN allows the construction of datasets with multi-property-biased distributions tailored to specific application scenarios, facilitating the discovery of novel optoelectronic materials.

  • Data-Driven Parametrization of All-Atom Force Fields for Organic Semiconductors

    by Guojiang Zhao, Taiping Hu, Yingfeng Zhang, Hongshuai Wang, Lirong Wu, Shuqi Lu, FengTianCi Yang, Shangqian Chen, Zhifeng Gao, Xinyan Wang, et al.

    Organic semiconductors (OSCs) composed of π conjugated molecules have attracted significant interest in studying bulk properties such as molecular arrangements and electron mobility. However, current traditional force fields (FFs) offer limited torsion types, failing to cover the full chemical space of π conjugated molecules and hindering further molecular dynamics simulation in deducing bulk properties through statistical mechanics. In this study, we introduce OSCFF, a GAFF2-compatible FF that supports diverse torsions for conjugated molecules and enables high-accuracy RESP charge prediction through a neural network (NN). To develop the OSCFF, we construct two expansive and diverse molecular data sets: one consists of around 56,000 fragment geometries with torsion profiles and another consists of around 472,000 optimized molecular geometries with RESP charges. Using these data sets, we train NN models to predict RESP charges and fit the missing dihedral parameters in GAFF2 through automatic differentiation techniques. We further demonstrate that OSCFF achieves high accuracy in predicting torsional energy profiles, RESP charges, and radial distribution functions for conjugated systems. Additionally, we release the data sets, dihedral parameters, and RESP model as open-source resources. We believe OSCFF will serve as a valuable tool for advancing the study of bulk properties in OSCs.


  • Open Access

    Virtual Characterization via Knowledge-Enhanced Representation Learning: from Organic Conjugated Molecules to Devices

    by Guojiang Zhao, Qi Ou, Zifeng Zhao, Shangqian Chen, Haitao Lin, Xiaohong Ji, Zhen Wang, Hongshuai Wang, Hengxing Cai, Lirong Wu, et al.

    The rational design of organic functional devices relies on understanding structure–property–performance relationships through multi-scale characterization. However, traditional characterizations are costly and require multidisciplinary expertise. Here we present OCNet, a domain-knowledge-enhanced representation learning framework that, for the first time, enables unified virtual characterization from molecules to devices. Pre-trained on over ten million self-generated conjugated molecules and dimers, OCNet learns generalizable microscopic representations comparable to expert-crafted features. As a result, it surpasses state-of-the-art models by over 20% in predicting key computed and experimental molecular optoelectronic properties. OCNet further provides the first transferable model for predicting transfer integrals in thin films, enabling accurate mesoscale carrier mobility estimation via multiscale simulations. By integrating tight -binding-level electronic descriptors, OCNet achieves real-time, accurate prediction of device power conversion efficiency. Together, OCNet offers a unified and scalable foundation for virtual characterization of organic materials across multiple scales, with broad applicability in photovoltaics, displays, and sensing

  • Open Access

    Data-Driven Parametrization of All-atom force fields for Organic Semiconductors

    by Guojiang Zhao, Taiping Hu, Hongshuai Wang, Lirong Wu, Shuqi Lu, FengTianCi Yang, Shangqian Chen, Zhifeng Gao, Xinyan Wang, Zheng Cheng

    Organic semiconductors (OSCs) composed of π conjugated molecules have gained significant interest in the study of bulk properties such as molecular arrangement and electron mobility. However, the types of torsion in the traditional force field (FF) are limited, fail to cover the chemical space of π conjugated molecules, and thus, hinder further molecular dynamics (MD) simulation to deduce these bulk properties through statistical mechanics. In this study, we introduce OSCFF, which supports various types of torsion for conjugated molecules and enables the generation of RESP charges with high accuracy through the neural network (NN). To develop the OSCFF, we construct two expansive and highly diverse molecular datasets: one consists of around 56,000 fragment geometries with torsion profiles, and another consists of around 472,000 optimized molecule geometries with RESP charges. The OSCFF demonstrates high accuracy in predicting torsional energy profiles, RESP charges, and the radial distribution function (RDF) for conjugated molecule systems. Furthermore, our OSCFF is compatible with the GAFF2 and a pipeline is provided for automatically generating the Gromacs supported topology file. We expect OSCFF will reduce the manual effort required for MD simulations of OSCs and serve as a valuable tool for multiple stages of OSCs design.