| Interface residue pair prediction | | ComplexContact; RaptorX-Contact; RaptorX-Property; Gremlin; DNCON2; PSICOV; FreeContact; LSTM; LSTM with Graph Representation | Direct evolutionary coupling analysis (DCA), machine learning and deep learning methods | Interfacial residue pair prediction can help subsequent protein complex structure predictions, such as docking.Protein contact map prediction can help reconstruct the three-dimensional structure of protein complexes. | The accuracy of interface residues for prediction needs to be improved. |
| Protein structure prediction | Template-free | ZDOCK; GRAMM; DOT; SmoothDock; ClusPro; MolFit; FTDock; 3D-Dock; PIPER; pyDock; HDOCK; SDOCK; HEX; FRODOCK; InterEvDock; MDockPP; CoDockPP; HSYMDOCK; SAM; RosettaDock; ICM-DOCK; HADDOCK; ATTRACT; DARWIN; Multi-LZerD; AutoDock | The search strategies of these methods are mainly FFT, GA and MC. | Protein docking can give all possible complex structures, some of which can also dock Cn and Dn complexes. | Designing an effective scoring function to sort the docking structure remains to be further explored. |
| Template-based | InterPreTS; Multimeric threading approach; M-Tasser; PISA; ProtCID | Using sequence or structure similarity to model protein complexes with known structures. | Template-based methods mainly reduce the possible structure by restricting the direction of protein binding. This method is more efficient than docking and can be applied to larger-scale protein complex prediction. | For proteins without a template, the structure of the complex cannot be predicted. |
| Protein complex prediction from PPI networks | Complex prediction based on PPI network clustering | MCODE; MCL; SPC; LCMA; SuperComplex; BN; CFinder; DPClus; IPCA; CMC; ClusterONE; HACO | The protein complex is part of a known PPI network, that is, the graph composed of protein complexes and their interactions is a subgraph of the PPI network. | Some of these methods only use the PPI network for clustering, and some use additional biological information, including structure, function, organization and co-evolution infornation, etc. | The proteins that may form complexes can only be picked out from the existing PPI network. |
| Complex interaction link prediction from PPI network | GGA; HAC; ECT; RWR; MDS; Link-weighted PPI | Methods to predict actual links in the network include public neighbors-based methods and distance-based methods. | This type of method predicts possible protein--protein interactions based on existing network information. | Public neighbors-based methods have limited effect on sparse networks. |
