Table 1.
ML-based software/model used for drug discovery.
| Name | Algorithm | Specific Function | PMID |
|---|---|---|---|
| Prediction of the target protein structure | |||
| TrRosetta server | DNN | Predict 3D structures of proteins | [13] |
| AlphaFold | DNN | Predict 3D structures of proteins | [14] |
| ComplexQA | GNN | Predict protein complex structure | [15] |
| ProteinBERT | Transformer | Predict secondary structure | [16] |
| ESMfold | Transformer | Predict structure of proteins | [17] |
| Predicting protein–protein interactions | |||
| IntPred | RF | Predict PPI interface sites | [18] |
| eFindSite | SVM; NBC | Predict PPI interfaces | [19] |
| DELPHI | RNN; CNN | Predict PPI sites | [20] |
| PPISP-XGBoost | XGBoost | Predict PPI sites | [21] |
| HN-PPISP | CNN | Predict PPI sites | [22] |
| TAGPPI | GCN | Predict PPIs | [23] |
| Struct2Graph | GAT | Predict PPIs | [24] |
| DeepFE-PPI | DNN | Predict PPIs | [25] |
| SGPPI | GCN | Predict PPIs | [26] |
| DeepPPI | DNN | Predict PPIs | [27] |
| DL-PPI | GNN | Predict PPIs | [28] |
| DeepSG2PPI | CNN | Predict PPIs | [29] |
| MaTPIP | Transformer; CNN | Predict PPIs | [30] |
| ProtInteract | Autoencoder; CNN | Predict PPIs | [31] |
| Predicting drug–target interactions | |||
| DeepC-SeqSite | CNN | Predict DTI binding sites | [32] |
| DeepSurf | CNN; ResNet | Predict DTI binding sites | [33] |
| PrankWeb | RF | Predict DTI binding sites | [34] |
| PUResNet | ResNet | Predict DTI binding sites | [35] |
| AGAT-PPIS | GNN | Predict DTI binding sites | [36] |
| DeepDTA | CNN | Predict DTI binding affinity | [37] |
| SimBoost | GBM | Predict DTI binding affinity | [38] |
| DEELIG | CNN | Predict DTI binding affinity | [39] |
| DeepDTAF | CNN | Predict DTI binding affinity | [8] |
| GraphDelta | CNN | Predict DTI binding affinity | [40] |
| PotentialNet | CNN | Predict DTI binding affinity | [41] |
| DeepAffinity | RNN, CNN | Predict DTI binding affinity | [9] |
| TeM-DTBA | CNN | Predict DTI binding affinity | [42] |
| Wang et al.’s method | RL | Predict DTI binding pose | [43] |
| Nguyen et al.’s method | RF; CNN | Predict DTI binding pose | [44] |
| AMMVF-DTI | GAT; NTN | Predict drug–target interactions | [45] |
| De novo drug design | |||
| ReLeaSE | RNN; RL | Conduct de novo drug design | [46] |
| ChemVAE | CNN; GRU | Conduct de novo drug design | [47] |
| MolRNN | RNN | Conduct multi-objective de novo drug design | [48] |
| PaccMann(RL) | VAE | Generate compounds with anti-cancer drug properties | [49] |
| druGAN | AAE | Conduct de novo drug design | [50] |
| SCScore | CNN | Evaluate the molecular accessibility | [51] |
| UnCorrupt SMILES | Transformer | Conduct de novo drug design | [52] |
| PETrans | Transfer learning | Conduct de novo drug design | [53] |
| FSM-DDTR | Transformer | Conduct de novo drug design | [54] |
| DNMG | GAN | Conduct de novo drug design | [55] |
| MedGAN | GAN | Design novel molecule | [56] |
| Prediction of the physicochemical properties | |||
| Panapitiya et al.’s method | GNN | Predict aqueous solubility | [57] |
| SolTranNet | Transformer | Predict aqueous solubility | [58] |
| Zang et al.’s method | SVM | Predict multiple physicochemical properties | [59] |
| Prediction of the ADME/T properties | |||
| ADMETboost | XGBoost | Predict ADME/T properties | [60] |
| vNN | k-NN | Predict ADME/T properties | [61] |
| Interpretable-ADMET | CNN; GAT | Predict ADME/T properties | [62] |
| XGraphBoost | GNN | Predict ADME/T properties | [63] |
| DeepTox | DNN | Predict toxicity of compounds | [64] |
| Li et al.’s method | DNN | Predict human Cytochrome P450 inhibition | [65] |
| LightBBB | LightGBM | Predict blood–brain barrier | [66] |
| Deep-B3 | CNN | Predict blood–brain barrier | [67] |
| PredPS | GNN | Predict stability of compounds in human plasma | [68] |
| Khaouane et al.’s method | CNN | Predict plasma protein binding | [69] |
| Application of AI in drug repurposing | |||
| deepDTnet | Autoencoder | Predict new targets of known drugs | [70] |
| NeoDTI | GCN | Predict new targets of known drugs | [71] |
| DTINet | Network diffusion algorithm and the dimensionality reduction | Predict new targets of known drugs | [72] |
| MBiRW | Birandom walk algorithm | Predict new indications of known drugs | [73] |
| GDRnet | GNN | Predict new indications of known drugs | [74] |
| deepDR | VAE | Predict new indications of known drugs | [75] |
| GIPAE | VAE | Predict new indications of known drugs | [76] |
| DrugRep-HeSiaGraph | Heterogeneous siamese neural network | Predict new indications of known drugs | [77] |
| iEdgeDTA | GCNN | Predict DTI binding affinity | [78] |
| Retrosynthesis prediction | |||
| Segler et al.’s method | MCTS, DNN | Predict retrosynthetic analysis | [79] |
| Liu et al.’s method | RNN | Predict retrosynthetic analysis | [80] |
| RAscore | RF | Predict retrosynthetic accessibility score | [81] |
| Reaction prediction | |||
| Wei et al.’s method | Neural network | Predict reaction classes | [82] |
| Coley et al.’s method | Neural network | Predict products of chemical reactions | [83] |
| Gao et al.’s method | Neural network | Predict optimal reaction conditions | [84] |
| Marcou et al.’s method | RF | Evaluate reaction feasibility | [85] |
Note: DNN, deep neural network; RNN, recurrent neural network; RF, random forest; CNN, convolutional neural network; XGBoost, eXtreme gradient boosting; GCN, graph convolutional network; GAT, graph attention network; SVM, support vector machine; NBC, naïve Bayes classifier; ResNet, residual network; GBM, gradient boosting machines; RL, reinforcement learning; GRU, gated recurrent unit; VAE, variational autoencoder; AAE, adaptive adversarial autoencoder; GNN, graph neural networks; k-NN, k-nearest neighbor; LightGBM, light gradient boosting machine; MCTS, Monte Carlo tree search, NTN, neural tensor network; GAN, generative adversarial network; GCNN, graph convolutional neural network.