机器学习的打分方程集合
受到机器学习（Machine Learning）和深度学习（Deep Learning）等算法模型的创新性冲击，其应用范围涵盖了自然语言处理（Natural Language Processing）、自动驾驶（Autopilot）、金融分析（Financial analysis）和生物医药研发（Biopharmaceutical R&D）等多个领域。其中，应用深度学习算法在预测小分子药物与疾病靶点蛋白之间的结合亲和力方面的设计和应用变得日益重要。分子对接(Molecular Docking)计算中打分方程(Scoring Function)算法通常用于对药物活性数据进行评估、分类、排名或预测，其结果对于决策制定候选药物分子而言至关重要。因此，本文将介绍一些基于机器学习和深度学习的打分方程。
1. 亲和力预测模型
| Model | Algorithm | Rp | RMSE | Reference |
|---|
| AGL-Score | GBT | 0.83 | 1.27 | [13] |
| ECIF | GBT | 0.87 | 1.17 | [14] |
| AEScore | MLP | 0.83 | 1.22 | [15] |
| OnionNet-2 | CNN | 0.86 | 1.16 | [16] |
| graphDelta | GNN | 0.87 | 1.05 | [17] |
| PointTransformer | CNN+ATT | 0.85 | 1.19 | [18] |
2. 机器学习打分方程
| Scoring Function | Algorithm | Description of protein–ligand complexes | Reference |
|---|
| RF-Score | RF | Protein−ligand atom-type pair counts | Ballester et al. (2010) |
| NN-Score 2.0 | ANN | Autodock Vina interaction terms, protein−ligand atom-type pair counts and electrostatic terms (BINANA) | Durrant and McCammon (2011) |
| ID-Score | SVM | Nine categories of descriptors related to protein–ligand interactions | Li et al. (2013) |
| SFCscore R F ^{RF} RF | RF | SFCscore interaction terms | Zilian and Sotriffer (2013) |
| ΔVinaRF 20 _{20} 20 | RF | Autodock Vina interaction terms and additional molecular descriptors | Wang and Zhang (2017) |
| RI-Score | RF | Rigidity index descriptors | Nguyen et al. (2017) |
| TNet-BP | CNN | Algebraic topology | Cang and Wei (2017) |
| K D E E P _{DEEP} DEEP | CNN | Molecular descriptors embedded into a 3D grid | Jiménez et al. (2018) |
| TopBP-ML | GBT | Algebraic topology | Cang et al. (2018) |
| TopBP-DL | CNN | Algebraic topology | Cang et al. (2018) |
| Pafnucy | CNN | Molecular descriptors embedded into a 3D grid | Stepniewska-Dziubinska et al. (2018) |
| PLEC-nn | DNN | Hashed fingerprint constructed by pairing ligand and protein atoms according to its environment | Wójcikowski et al. (2019) |
| EIC-Score | GBT | Differential geometry representations | Nguyen and Wei (2019b) |
| AGL-Score | GBT | Statistical features of the adjacency and Laplacian matrices of multiscale weighted labeled algebraic subgraphs | Nguyen and Wei (2019a) |
| OnionNet | CNN | Rotation-free element pair-specific contacts between ligands and protein atoms, grouped into different distance ranges | Zheng et al. (2019) |
| ΔVinaXGB | XGBT | Autodock Vina score and molecular descriptors, including water molecules | Lu et al. (2019) |
| NNScore::LD | FFNN | NNScore 2.0 features and RDKit ligand descriptors | Boyles et al. (2020) |
| RosENet | CNN | Molecular mechanics energies from Rosetta force field and molecular descriptors embedded onto a 3D grid | Hassan-Harrirou et al. (2020) |
| ECIF-GBT | GBT | Protein−ligand atom-type pair counts considering each atoms connectivity | Norberto et al. (2021) |
| ECIF::LD-GBT | GBT | Protein−ligand atom-type pair counts considering each atoms connectivity and RDKit ligand descriptors | Norberto et al. (2021) |
