Table 1 Classification of deep-learning PLBAP models
From: Structure-based, deep-learning models for protein-ligand binding affinity prediction
Type | Feature representation \({\mathcal {R}}\) | Symmetry properties\(^*\) of \({\mathcal {R}}\) | Key learning architecture | Model interpretability | Representatives |
|---|---|---|---|---|---|
\(T_{ACNN}\) | Atom coordinates & types | TE/RE/PE | Concatenated ACNNs | Model-level | ACNN [25] |
\(T_{IMC-CNN}\) | IMC profiles | TI/RI/PI | 2D-CNNs | None | |
\(T_{Grid-CNN}\) | Grid voxels | TI/RE/PI | 3D-CNNs | Post-hoc analysis | KDEEP [29], Pafnucy [38], CNN-Score [39], DeepAtom [40], Sfcnn [41] |
\(T_{Graph-GCN}\) | Molecular graphs | TI/RI/PI | GCNs | Model-level | GraphBAR [30], APMNet [42], PotentialNet [43], GraphDTI [44] |