Skip to main content

Table 1 Performance comparison of various methods on LBA dataset under different protein sequence identity split settings

From: Harnessing pre-trained models for accurate prediction of protein-ligand binding affinity

Method

LBA30

LBA60

RMSE \(\downarrow\)

Pearson \(\uparrow\)

Spearman \(\uparrow\)

RMSE \(\downarrow\)

Pearson \(\uparrow\)

Spearman \(\uparrow\)

DeepDTA [34]

1.866

0.472

0.471

1.762

0.666

0.663

TAPE [35]

1.890

0.338

0.286

1.633

0.568

0.571

ProtTrans [36]

1.544

0.438

0.434

1.641

0.595

0.588

Atom3D-CNN [33]

1.416

0.550

0.553

1.621

0.608

0.615

Atom3D-ENN [33]

1.568

0.389

0.408

1.620

0.623

0.633

Atom3D-GNN [33]

1.601

0.545

0.533

1.408

0.743

0.743

Holoprot [37]

1.464

0.509

0.500

1.365

0.749

0.742

ProNet [38]

1.463

0.551

0.551

1.343

0.765

0.761

DeepAffinity [39]

1.893

0.415

0.426

–

–

–

SMT-DTA [40]

1.574

0.458

0.447

1.347

0.758

0.754

GeoSSL [41]

1.451

0.577

0.572

–

–

–

Uni-Mol [27]

1.434

0.565

0.540

1.357

0.753

0.750

BindNet [28]

1.340

0.632

0.620

1.230

0.793

0.788

SableBind

1.527

0.579

0.579

1.246

0.802

0.798

Cross-Validation

1.581

0.616

0.620

1.562

0.630

0.632

  1. The methods are organized into three categories: sequence-based methods at the top, followed by structure-based methods, and finally pre-training approaches at the bottom. The best and second-best results are highlighted in bold and underlined, respectively. The results of SableBind-cross-validation are presented with a italic background, and are not included in the ranking for best results due to differing dataset partitioning
Back to article page

BMC Bioinformatics

ISSN: 1471-2105

Contact us