LiTENexus

An End-to-End Virtual Drug Discovery System Based on Quantum Chemical Grounding

Physics-informed molecular intelligence for next-generation AI-driven drug discovery

---

✨ Overview

LiTENexus is a physics-grounded, end-to-end AI-aided drug discovery (AIDD) framework designed to bridge microscopic quantum chemical principles with macroscopic pharmacological applications. The system is built upon the Quantum Chemical Information Injection (QCII) mechanism and powered by the foundational LiTEN-Base operator.

Unlike conventional molecular representation models that primarily learn statistical correlations from data, LiTENexus dynamically integrates quantum chemical laws into neural operators, enabling:

• 🔬 Enhanced out-of-distribution (OOD) generalization
• 🧠 Improved mechanistic interpretability
• ⚛️ Quantum-level microscopic property prediction
• 💊 High-performance ADMET prediction
• 🔗 Cross-modal virtual screening
• 🚀 End-to-end virtual drug discovery workflows

---

🌟 Key Features

⚛️ Quantum-Chemical Grounded Representation Learning

• Physics-informed neural operator architecture
• Dynamic injection of quantum chemical information
• Field reconstruction paradigm for spatial topology and polarization modeling
• Unified latent representation across molecular tasks

🧪 Microscopic Property Prediction

• Quantum-level accuracy comparable to DFT methods

• 2–3 orders of magnitude faster than conventional DFT calculations

• Support for:

  • Energy prediction
  • Force prediction
  • Charge distribution
  • Molecular conformational analysis

💊 ADMET Modeling

• State-of-the-art performance across multiple benchmarks

• Strong generalization to:

  • Natural products
  • Cyclic peptides
  • Complex out-of-distribution molecular spaces

• Multi-task pharmacokinetic property prediction

🔍 Virtual Screening

• Quantum Manifold Dense Retrieval (QMDR)
• Cross-modal molecular representation alignment
• Efficient ligand retrieval and ranking
• High enrichment performance on benchmark datasets

---

🧩 Modular Ecosystem

LiTENexus is composed of multiple interoperable modules:

Module	Description
LiTEN-Base	Foundational quantum-aware representation operator
LiTEN-FF	Molecular force field and conformational modeling
LiTEN-Micro	Microscopic physicochemical property prediction
LiTEN-ADMET	Pharmacokinetic and toxicity prediction
LiTENCLIP	Cross-modal molecular retrieval and screening

---

🏗️ System Architecture

               Microscopic Quantum Dynamics
                              │
                              ▼
      Quantum Chemical Information Injection
                         (QCII)
                              │
                              ▼
         LiTEN-Base Universal Representation Engine
                              │
       ┌──────────────┬──────────────┬──────────────┬
       ▼              ▼              ▼              ▼
      LiTEN-FF    LiTEN-Micro   LiTEN-ADMET    LiTENCLIP
       │              │              │              │
      Conformation  Physicochemical  ADMET      Virtual
      Modeling      Properties       Prediction  Screening

---

🚀 Highlights

• ⚛️ Quantum-chemically grounded molecular representation learning
• 🌍 Strong OOD generalization ability
• 🔄 Unified microscopic-to-macroscopic modeling pipeline
• 🧠 Physics-aware neural operator architecture
• 🔍 Scalable virtual screening framework
• 🧩 Modular and extensible design

---

📊 Benchmark Performance

💊 ADMET Prediction

• Outperformed more than 20 baseline models on proprietary datasets

• Achieved leading performance on:

  • PharmaBench
  • Biogen
  • ADMETLAB 3.0

🔍 Virtual Screening

Dataset	Performance
DUD-E	High enrichment performance
LIT-PCBA	Strong retrieval capability

⚛️ Quantum Property Modeling

• Near-DFT-level prediction accuracy
• Significant computational acceleration compared with traditional quantum chemistry methods

---

🧬 Applications

LiTENexus can be applied to:

• AI-aided drug discovery
• Molecular property prediction
• ADMET optimization
• Lead compound screening
• Molecular generation
• Conformation analysis
• Quantum chemistry acceleration
• Physics-informed molecular modeling

---

🌐 Online Platform

The public LiTENexus Platform is under continuous development and expansion, aiming to provide open access to quantum-aware molecular modeling and AI-driven drug discovery tools.

Current and upcoming functionalities include:

• ⚛️ Quantum-level molecular property prediction
• 💊 ADMET evaluation
• 🧪 Molecular conformation optimization
• 🔍 Cross-modal virtual screening
• 🔗 Molecular retrieval and ranking
• 🤖 Interactive AI-assisted drug discovery workflows

🔗 Platform Access

👉 Platform Website:
https://cadd.zju.edu.cn/litenexus/

The platform is continuously being updated with new models, datasets, and drug discovery modules. More features and public services will be released progressively.

---

📖 Citation

If you find this project useful in your research, please cite:

@article{
doi:10.26434/chemrxiv.15003380/v1,
author = {Qun Su  and Qiaolin Gou  and Hui Zhang  and Meijing Fang  and Kewen Wang  and Wangcong Tian  and Yurong Li  and Donghai Zhao  and Yitong Li  and Rui Qin  and Shicheng Chen  and Zijie Chen  and Peichen Pan  and Yu Kang  and Chang-Yu Hsieh  and Jike Wang  and Tingjun Hou },
title = {LiTENexus: An End-to-End Virtual Drug Discovery System Based on Quantum Chemical Grounding},
journal = {ChemRxiv},
volume = {2026},
number = {0515},
pages = {},
year = {2026},
doi = {10.26434/chemrxiv.15003380/v1},
URL = {https://chemrxiv.org/doi/abs/10.26434/chemrxiv.15003380/v1}
}

---

📬 Correspondence

• Tingjun Hou — tingjunhou@zju.edu.cn
• Jike Wang — jikewang@zju.edu.cn
• Chang-Yu Hsieh — kimhsieh@zju.edu.cn
• Yu Kang — yukang@zju.edu.cn

---

📄 License

This project is licensed under the Apache License 2.0.

---

🙏 Acknowledgements

We thank all collaborators and contributors involved in the development of LiTENexus and related quantum chemistry and AI-driven drug discovery research.

---

⭐ Star History

If you find this repository useful, please consider giving it a ⭐ on GitHub.