AI-Drug-Discovery-Assistant: An AI-Driven Molecular Docking Drug Discovery Platform Based on Agent Calls

AI-Drug-Discovery-Assistant is an AI-driven molecular docking drug discovery platform built on Spring Boot. It integrates the AutoDock Vina molecular docking engine and RDKit cheminformatics toolkit, supporting drug discovery workflows such as receptor preparation and ligand generation through intelligent tool-calling agents. Its goal is to lower the technical threshold for drug research and development and achieve intelligent automation of processes.

Traditional drug research and development is time-consuming (over a decade) and costly (billions of dollars). Molecular docking, as a core technology in structure-based drug design, is a key step in virtual screening and lead compound optimization. However, existing workflows involve complex computational processes and require high professional skills from researchers.

Core components include:

1. Spring Boot backend framework: Provides RESTful APIs, task scheduling, and data management, ensuring scalability and stability;
2. AutoDock Vina docking engine: An open-source molecular docking program using optimized scoring functions and search algorithms;
3. RDKit cheminformatics toolkit: Written in Python, supporting molecular descriptor calculation, substructure search, and other functions;
4. Intelligent tool-calling agent: The core innovation of the platform, coordinating components to complete complex workflows.

Receptor Preparation Module

Automatically processes PDB structure files (missing atom supplementation, hydrogen atom addition, etc.), and the intelligent agent recommends optimal parameters to reduce manual trial and error.

Ligand Generation and Management

Supports input methods such as database import, SMILES parsing, and scaffold generation, and integrates ADMET property prediction to identify druggability risks.

Intelligent Workflow Orchestration

Through the AI agent's understanding of natural language instructions, it automatically plans multi-step workflows (e.g., ACE2 receptor inhibitor screening).

Result Visualization and Analysis

Interactive 3D display of docking results, supporting binding mode analysis and SAR analysis tools.

Built based on large language models, it has:

• Intent understanding: Parses user natural language queries;
• Tool selection: Dynamically selects computational tools;
• Parameter optimization: Sets optimal parameters based on historical data;
• Error recovery: Diagnoses problems and tries alternative solutions;
• Result synthesis: Generates structured research reports.

Applicable scenarios:

• Virtual screening: Quickly identifies potential hit compounds;
• Lead optimization: Guides chemical modifications to improve affinity;
• Target validation: Evaluates the druggability of new targets;
• Drug repurposing: Discovers new indications for existing drugs;
• Teaching and training: Computational biology practice platform. The platform reflects the trend of integration between AI and scientific computing, lowering the technical threshold to accelerate research and development.

The roadmap includes:

1. Integrate molecular dynamics simulation tools to evaluate binding stability;
2. Introduce generative AI models for de novo molecular design;
3. Build a drug discovery knowledge graph to support intelligent reasoning;
4. Deeply integrate public databases such as ChEMBL and PubChem.