Protein-Ligand Binding Affinity Prediction on PDBbind Sequence Identity (30%) 2017
1.295RMSEh-MINT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| h-MINTEncoder Type=Joint Encoder2026.04 | 1.295 | 0.64 | 0.625 | |
| CT-SCD-SAIR-PocketPretraining=SCD, Parameters=10M2026.03 | 1.304 | — | — | |
| CT-SCD-SAIR-PocketCategory=SCD Pretrained, Params=10M2026.03 | 1.304 | 0.64 | 0.624 | |
| ADiT-LPretraining=Other, Parameters=253M2026.03 | 1.308 | — | — | |
| ADiT-LCategory=Other Pretrained, Params=253M2026.03 | 1.308 | 0.645 | 0.647 | |
| EPT-MultiPretraining=Other, Parameters=30M2026.03 | 1.322 | — | — | |
| EPT-MultiDomainCategory=Other Pretrained, Params=30M2026.03 | 1.322 | 0.644 | 0.63 | |
| EPT-ProteinPretraining=Other, Parameters=30M2026.03 | 1.329 | — | — | |
| EPT-ProteinCategory=Other Pretrained, Params=30M2026.03 | 1.329 | 0.628 | 0.613 | |
| CT-SCD-PCQPretraining=SCD, Parameters=10M2026.03 | 1.332 | — | — | |
| CT-SCD-PCQCategory=Pretrained on PCQ, Params=10M2026.03 | 1.332 | 0.617 | 0.6 | |
| EPT-MoleculeCategory=Other Pretrained, Params=30M2026.03 | 1.336 | 0.621 | 0.602 | |
| ADiT-SPretraining=Other, Parameters=12M2026.03 | 1.337 | — | — | |
| CT-SCD-SAIRPretraining=SCD, Parameters=10M2026.03 | 1.337 | — | — | |
| ADiT-SCategory=Other Pretrained, Params=12M2026.03 | 1.337 | 0.626 | 0.618 | |
| CT-SCD-SAIRCategory=SCD Pretrained, Params=10M2026.03 | 1.337 | 0.617 | 0.599 | |
| ADiT-MPretraining=Other, Parameters=35M2026.03 | 1.353 | — | — | |
| ADiT-MCategory=Other Pretrained, Params=35M2026.03 | 1.353 | 0.622 | 0.63 | |
| FradCategory=Pretrained on PCQ, Params=14M2026.03 | 1.365 | 0.599 | 0.577 | |
| CT-SCD-ALLPretraining=SCD, Parameters=10M2026.03 | 1.372 | — | — | |
| CT-SCD-ALLCategory=SCD Pretrained, Params=10M2026.03 | 1.372 | 0.594 | 0.578 | |
| ProFSACategory=Other Pretrained, Params=47.6M2026.03 | 1.377 | 0.628 | 0.62 | |
| EPT-ScratchPretraining=None, Parameters=30M2026.03 | 1.378 | — | — | |
| EPT-ScratchCategory=No Pretraining, Params=30M2026.03 | 1.378 | 0.604 | 0.594 | |
| GET-PSEncoder Type=Joint Encoder2026.04 | 1.387 | 0.601 | 0.582 | |
| CT-SCD-OMOL25Pretraining=SCD, Parameters=10M2026.03 | 1.389 | — | — | |
| CT-SCD-OMOL25Category=SCD Pretrained, Params=10M2026.03 | 1.389 | 0.586 | 0.571 | |
| CT-FE-OMOL25Category=FE Pretrained, Params=10M2026.03 | 1.391 | 0.575 | 0.564 | |
| CT-SCD-GEOM10Category=SCD Pretrained, Params=10M2026.03 | 1.392 | 0.606 | 0.554 | |
| EGNN-PLMCategory=Other Pretrained, Params=650M2026.03 | 1.403 | 0.565 | 0.544 | |
| CT-SCD-AMP20Category=SCD Pretrained, Params=10M2026.03 | 1.408 | 0.584 | 0.533 | |
| GET-BRICSEncoder Type=Joint Encoder2026.04 | 1.41 | 0.592 | 0.579 | |
| CT-SCD-SAIR-LigCategory=SCD Pretrained, Params=10M2026.03 | 1.412 | 0.566 | 0.547 | |
| GET-MurckoEncoder Type=Joint Encoder2026.04 | 1.415 | 0.59 | 0.578 | |
| ATOM3D-3DCNNCategory=No Pretraining, Params=—2026.03 | 1.416 | 0.55 | 0.553 | |
| Atom3D-3DCNNEncoder Type=Joint Encoder2026.04 | 1.416 | 0.55 | 0.553 | |
| Townshend et al. (ENN)Architecture=Equivariant neural network (ENN), Scope=Binding pocket only, Category=Structure-based Methods2022.04 | 1.429 | 0.541 | 0.532 | |
| GETEncoder Type=Joint Encoder2026.04 | 1.43 | 0.586 | 0.575 | |
| RicciBind2026.06 | 1.437 | 0.622 | — | |
| EHIGN2026.06 | 1.449 | 0.613 | — | |
| ProNet-Amino AcidEncoder Type=Separate Encoder2026.04 | 1.455 | 0.536 | 0.526 | |
| ProNet-BackboneEncoder Type=Separate Encoder2026.04 | 1.458 | 0.546 | 0.55 | |
| ProtNetPretraining=None2026.03 | 1.463 | — | — | |
| ProtNet-All-AtomCategory=No Pretraining, Params=—2026.03 | 1.463 | 0.551 | 0.551 | |
| ProNet-All-AtomEncoder Type=Separate Encoder2026.04 | 1.463 | 0.551 | 0.551 | |
| HOLOPROT (full surface)# Params=1.44 M, Configuration=full surface2022.04 | 1.464 | 0.509 | 0.5 | |
| HoloProt-Full SurfaceCategory=No Pretraining, Params=1.4M2026.03 | 1.464 | 0.509 | 0.5 | |
| Holoprot-Full SurfaceEncoder Type=Separate Encoder2026.04 | 1.464 | 0.509 | 0.5 | |
| RF-Score2026.06 | 1.469 | 0.582 | — | |
| EGNN2026.06 | 1.483 | 0.579 | — | |
| Gainza et al.# Params=0.62 M, Category=Surface-based Methods2022.04 | 1.484 | 0.467 | 0.455 | |
| MaSIFEncoder Type=Separate Encoder2026.04 | 1.484 | 0.467 | 0.455 | |
| HOLOPROT (molecular superpixels)# Params=1.76 M, Configuration=molecular superpixels2022.04 | 1.491 | 0.491 | 0.482 | |
| Holoprot-SuperpixelEncoder Type=Separate Encoder2026.04 | 1.491 | 0.491 | 0.482 | |
| GNN-DTI2026.06 | 1.499 | 0.562 | — | |
| SchNet2026.06 | 1.499 | 0.566 | — | |
| MetalProGNet2026.06 | 1.5 | 0.574 | — | |
| IGN2026.06 | 1.509 | 0.561 | — | |
| CT (Baseline)Pretraining=None, Parameters=10M2026.03 | 1.51 | — | — | |
| CTCategory=No Pretraining, Params=10M2026.03 | 1.51 | 0.501 | 0.486 | |
| Uni-MolPretraining=Other, Parameters=47.6M2026.03 | 1.52 | — | — | |
| Uni-MolCategory=Other Pretrained, Params=47.6M2026.03 | 1.52 | 0.558 | 0.54 | |
| GIGN2026.06 | 1.521 | 0.565 | — | |
| ESM-2 + fingerprintEncoder Type=Separate Encoder2026.04 | 1.537 | 0.455 | 0.433 | |
| SS-GNN2026.06 | 1.539 | 0.573 | — | |
| Elnaggar et al.# Params=2.4 M, Category=Sequence-based Methods, Fine-tuning=No (embeddings saved to disk)2022.04 | 1.544 | 0.438 | 0.434 | |
| ProtTransEncoder Type=Separate Encoder2026.04 | 1.544 | 0.438 | 0.434 | |
| Hermosilla et al.# Params=5.80 M, Category=Structure-based Methods2022.04 | 1.554 | 0.414 | 0.428 | |
| IEConvEncoder Type=Separate Encoder2026.04 | 1.554 | 0.414 | 0.428 | |
| ATOM3D-ENNCategory=No Pretraining, Params=—2026.03 | 1.568 | 0.389 | 0.408 | |
| Atom3D-ENNEncoder Type=Joint Encoder2026.04 | 1.568 | 0.389 | 0.408 | |
| PotentialNet2026.06 | 1.582 | 0.507 | — | |
| GVPEncoder Type=Joint Encoder2026.04 | 1.594 | — | — | |
| ATOM3D-GNNCategory=No Pretraining, Params=—2026.03 | 1.601 | 0.545 | 0.533 | |
| Atom3D-GNNEncoder Type=Joint Encoder2026.04 | 1.601 | 0.545 | 0.533 | |
| Öztürk et al.# Params=1.93 M, Category=Sequence-based Methods2022.04 | 1.866 | 0.472 | 0.471 | |
| DeepDTAEncoder Type=Separate Encoder2026.04 | 1.866 | 0.472 | 0.471 | |
| Rao et al.# Params=93.0 M, Category=Sequence-based Methods2022.04 | 1.89 | 0.338 | 0.286 | |
| TAPEEncoder Type=Separate Encoder2026.04 | 1.89 | 0.338 | 0.286 | |
| Townshend et al. (GNN)Architecture=Graph neural network (GNN), Scope=Binding pocket only, Category=Structure-based Methods2022.04 | 1.936 | 0.581 | 0.647 | |
| Bepler and Berger# Params=48.8 M, Category=Sequence-based Methods2022.04 | 1.985 | 0.165 | 0.152 | |
| Bepler and Berger'sEncoder Type=Separate Encoder2026.04 | 1.985 | 0.165 | 0.152 |