Overall Structure

The codebase is organized into several major components: 1 Core Library (src/lib/): Contains fundamental data structures and algorithms 2 CUDA Implementation (src/cuda/): GPU-accelerated components for performance 3 Main Application (src/main/): Command-line interface and application entry point 4 Split Utility (src/split/): Tool for processing multi-model PDBQT files

Workflow

A typical Uni-Dock workflow appears to be: 1 Parse input files (receptors and ligands) 2 Set up the scoring function and parameters 3 Generate or load grid maps for efficient scoring 4 Perform global search (docking) 5 Refine and score the resulting poses 6 Output the best conformations

File Format

• PDBQT是AutoDock和AutoDock Vina等分子对接软件专用的格式，用于描述受体（如蛋白质）和配体（小分子）的结构信息。它在PDB格式基础上扩展了电荷（Q） 和原子类型（T） 两个关键字段

• SDF（Structure-Data File）是化学信息学中通用的分子结构描述格式，主要用于存储小分子的2D/3D结构、化学性质及实验数据 （如生物活性）。常见于PubChem、DrugBank等数据库

• unidock/src/lib/parse_pdbqt.cpp

• unidock/src/split/split.cpp：处理 Multi-model PDBQT 多构象分子

  • 通过将每个模型提取到单独的文件中，以便在实际对接过程中更有效地处理。

unidock/src/main/main.cpp

1 Command Line Interface: Uses Boost’s program_options library to process command-line arguments and configuration files.

• Input: Specifies receptors, flexible residues, and ligand files
• Search space: Defines the docking grid dimensions and positioning
• Output: Controls where results are stored
• Advanced options: Fine-tunes the docking algorithm behavior
• Misc options: Sets general parameters like CPU usage and random seed

Boost是C++领域最具影响力的开源程序库集合，由全球开发者社区共同维护，旨在为C++标准库提供扩展功能，被称为“准标准库”

残基 （Residue）是生物大分子（如蛋白质、核酸）中构成其基本单元的组成部分，由氨基酸或核苷酸通过化学键连接后剩余的结构片段形成。 分子对接中的柔性残基 （Flexible Residues）是指在对接过程中允许构象变化的受体蛋白残基。这类残基通常位于结合口袋内，其侧链或主链的灵活性会影响配体与受体的结合模式。通过模拟这些残基的动态调整，柔性残基对接能够更真实地反映生物分子间的相互作用，提升预测准确性。

2 Docking Workflow Management: Handles various docking workflows including:

• Single ligand-receptor docking on CPU
• Batch docking of multiple ligands on GPU
• Paired batch processing for one-to-one ligand-protein docking

3 Memory Management: Implements GPU memory management strategies to optimize batch processing.

A sophisticated batch processing system: 1 Classifies ligands into size categories (small, medium, large, extra large) 2 Processes ligands in batches to maximize GPU utilization 3 Predicts memory requirements to determine optimal batch sizes