Recently，At DP Technology2025At the Developer Conference,DP Technology andDongyang Light Medicine Joint ReleaseThe World's First Pharmacokinetics (PK) Prediction Model Based on Pre-training and Neural Ordinary Differential Equations (Neural ODE) CouplingThis breakthrough technology, by deeply integrating mechanism modeling with deep learning, has achieved precise prediction of drug concentration-time curves in rat experimental data. The coefficient of determination (R²) for the top 30% predicted samples is as high as 0.891, significantly outperforming traditional PK models.This innovative model has been successfully applied to the preclinical R&D pipeline of Dongyangguang Pharmaceutical. By dynamically simulating the drug absorption, distribution, metabolism, and excretion (ADME) processes, it achieves the following core values:1. Mechanism Interpretability: Retain the physical description of physiological processes using nonlinear ODEs to ensure that prediction results conform to pharmacological laws.2. Improved Prediction Accuracy: UtilizePre-trained Models and Neural ODE Solving Methods, improve the prediction accuracy of the PK model3. Multi-scenario Adaptation: Support prediction inUnder different administration routesConcentration-Time Curve (CT Curve) of the DrugThis technology marks the evolution of AI from an auxiliary analysis tool to a core decision-making engine.Pioneering a closed-loop paradigm of "experimental data-mechanistic modeling-intelligent prediction" for innovative drug development, which is expected to drive the pharmaceutical industry's transition from "trial-and-error development" to a "precision design" model.。Dongyang Light PK Model Framework DiagramUni-PK Model:
Scarce Experimental Data+AlgorithmGenerational Leap
1. Exclusive Data Moat: How HEC Pharm Defines the Next Competitive Dimension of AI Drug DiscoveryAlthough numerous powerful models have emerged in the AI field, the shortage of high-quality and diverse training data remains a core bottleneck constraining their practical implementation. Public literature data often exhibits publication bias, with many studies tending to publish only positive results or findings in popular fields. Therefore, applying AI models to key scenarios in drug development requires breaking the "data curse."Based on this, DP Technology andHEC Pharm Jointly Launches New Uni-PK Model:- Construction of High-Quality Data Ecosystem: Relying on the innovative drug pipeline accumulation of Dongyang Guangyao, establish a PK database covering multiple species and dosage forms, and construct a high signal-to-noise ratio training set through systematic correction.
- Industry Data Fine-tuning Strategy: Adopt the paradigm of "general large model pre-training + pharmaceutical domain data fine-tuning" to inject prior knowledge of the pharmaceutical field while retaining AI generalization capabilities. Compared with models trained purely on public data, the prediction accuracy is greatly improved (internal validation data).
2. Uni-Mol Pre-trained Model: From Molecular Structure to PK PredictionMedicinalAbsorption, Distribution, Metabolism, and Excretion (ADME) ProcessIs largely influenced by itsMolecular Structure, Physicochemical Properties, and Biological InteractionsImpact. Relying solely on traditional parametric PK models or experimental data for modeling may not fully capture molecular-level characteristics. Therefore, this model innovativelyIntroducing Uni-MolPre-trained ModelUni-Mol is a product developed by DP Technology.Universal Molecular Pre-trained Model for Small Molecule Tasks, it is able toUtilizing Large-Scale Molecular Data for Self-Supervised Learning, and performs well in different molecular modeling tasks (such as molecular property prediction, protein-ligand interaction modeling, etc.). The application of this pre-trained model, on the one handEnhanced the interpretability of molecular information for PK curve prediction, making itCan beOptimizing PK Parameters Using Molecular Prior Knowledge in the Absence of Experimental Data; on the other handEnabling data to be utilized efficiently，Be able toEnhancing Molecular Dynamics Prediction Capability Under Small Sample Conditions, enabling the model to adapt to a wider range of compound types.'%20fill='%23FFFFFF'%3E%3Crect%20x='249'%20y='126'%20width='1'%20height='1'%3E%3C/rect%3E%3C/g%3E%3C/g%3E%3C/svg%3E)
Uni-Mol Framework Diagram
3. Neural ODE Solution: Deciphering the Complexity of Pharmacokinetics (PK)PK Curve: A curve describing the drug's behavior in the body during the drug development process.Indicators of Absorption, Distribution, Metabolism, and Excretion (ADME) Processes, in the algorithm its dynamic behavior is usually represented byCharacterization of Nonlinear Ordinary Differential Equations (ODEs)。However, in practical applications, PK systems often involve complex nonlinear characteristics such as saturated metabolism, time-varying parameters, and inter-individual variability.，Making it difficult for traditional analytical or semi-analytical methods to accurately model.This model adopts deep learning to optimize Neural ODE parameters, enabling the model toAutomatically Learn the Key Dynamic Characteristics of the PK System，Rat experimental data show that the R² of the Top 60% samples reaches 0.764, and the R² of the Top 30% high-quality samples jumps to 0.891.. Without strictly relying on preset kinetic structures, it not only improves the modeling capability of nonlinear PK processes but also enhancesGeneralization of the Model, making it applicable to a wider range of drug types and dosage regimens.
Predictive Performance of C-t Curve on the Test Set
Industrial Applications and Strategic Value
- R&D Paradigm Innovation: Since the implementation of the AI+ strategy in 2024, Dongyang Light Medicine has built a full-process AI R&D system covering target prediction, hit compound screening, lead compound optimization, and PK modeling. As a core module, the Uni-PK model is expected to achieve results in IND-stage projects such as HEC169584:
· Optimization cycle for candidate compounds reduced by 50% (from 12 weeks → 6 weeks)· Reduction of animal experiments by 60% (through screening and elimination of low-potential molecules)
- Ecosystem Construction: This model has been deployed on DP Technology's Bohr Platform, supporting pharmaceutical companies in conducting virtual PK experiments. Users can access it through [https://bohrium.dp.tech/apps/uni-pk]Apply for a trial to experience the end-to-end prediction service from "molecular structure → PK curve."
Task Submission Interface
Task Result Interface
About Dongyang Light Medicine
Established in 2003, Dongyang Guangya Pharmaceutical Co., Ltd. (HEC Pharm) is a comprehensive pharmaceutical company driven by independent research and development, rooted in China, and oriented towards the global market. HEC Pharm provides innovative, high-quality, and affordable medicines to patients worldwide, with the corporate vision of becoming a world-class pharmaceutical enterprise. Over the past 20 years, the company has adhered to the development strategies of "innovation" and "internationalization," establishing leading drug R&D platforms in China, production facilities that meet international standards, and a sales network covering the globe. The company strategically focuses on therapeutic areas such as pediatric medicines, infectious diseases, chronic conditions, and oncology. It has 150 approved drugs globally, over 100 drugs in the research pipeline, including nearly 50 first-class innovative drugs, gradually evolving into one of the leading enterprises in China's biopharmaceutical industry. For seven consecutive years, the company has been listed in the TOP20 of the "Comprehensive Ranking of China's Pharmaceutical R&D Strength" published by Pharma Intelligence, and ranked in the top tier of the "Top 100 Innovative Pharmaceutical Enterprises in China 2023." Its number of patents ranked 44th globally and 4th in China on the "TOP 100 Global Biomedical Industry Invention Patents Ranking 2022." According to the Frost & Sullivan report, based on the number of patents published and authorized patent announcements in China from January 1, 2014, to December 31, 2023, HEC Pharm ranked first among pharmaceutical companies in China.DP Technology is globallyAIFor Science Leaders, AI for Science means using AI to learn a series of scientific principles and knowledge, and further solve key problems in scientific research and industrial R&D fields.Relying on deep cultivation in the field of interdisciplinary research, DP Technology has built the "Deep Potential·Universe" AI for Science large model system, further addressing key issues in scientific research and industrial development, and bringing research methods across many disciplines from the "trial-and-error/experimentation" era into the "Pre-trained ModelThe "era" has formed the "innovation-implementation" chain and open ecosystem of AI for Science, built a micro-scale industrial infrastructure based on AI for Science, empowered "thousands of industries," and created a new generation of R&D systems for the most fundamental areas of human economic development: biomedicine, energy, materials, and information science and engineering research.DP Technology is a national high-tech enterprise in China.Specialized, Precise, Unique, and Innovative"Little Giant" enterprises have set up R&D centers in cities such as Beijing, Shanghai, and Shenzhen.The scientific research and technology team is led by an academician of the Chinese Academy of Sciences, bringing together over a hundred outstanding young scientists and engineers from various fields such as mathematics, physics, chemistry, biology, materials, and computer science. Among the company's members, more than 35% hold a Ph.D. or are postdoctoral researchers.Core members have won the "Gordon Bell Prize," the highest award in the field of global high-performance computing in 2020. Their related work was selected as one of China's top ten scientific and technological advances in 2020 and one of the ten major technological breakthroughs in the global AI field.
