Protein Property Prediction on 21 protein property datasets 48 data points (Cross-validation)
6.65Spearman CorrelationKermutSeq-GP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| KermutSeq-GPtype=Sequence2026.06 | 6.65 | 6.27 | 6.06 | |
| Ridge-OHRepresentation=One-Hot2026.06 | 6.54 | 6.9 | 6.9 | |
| SigGLM-OHRepresentation=One-Hot2026.06 | 6.03 | 5.6 | 5.59 | |
| Tanimoto-GP2026.06 | 5.89 | 5.75 | 5.81 | |
| Ridge-ESM2Backbone=ESM22026.06 | 5.16 | 5.03 | 5.22 | |
| KermutStruc-GPtype=Structure2026.06 | 4.73 | 5.06 | 5.67 | |
| Kermut-GP2026.06 | 4.05 | 4.19 | 4.33 | |
| MLP-ESM2-LastLayerBackbone=ESM2, Layer=LastLayer2026.06 | 3.87 | 4.24 | 3.49 | |
| LOCK-GP2026.06 | 2.08 | 1.95 | 1.92 | |
| LOCK-GPModel Class=Gaussian process, Prior Information=BLOSUM, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.655 | 0.682 | 0.496 | |
| LOCK-GP2026.06 | 0.655 | 0.682 | 0.496 | |
| Kermut-GPModel Class=Gaussian process, Prior Information=ESM-2 + ProteinMPNN, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.638 | 0.639 | 0.547 | |
| Kermut-GP2026.06 | 0.638 | 0.639 | 0.547 | |
| ConFit2026.06 | 0.628 | 0.615 | 0.574 | |
| Ridge-ESM2-650MBackbone=ESM2-650M, Classifier=Ridge2026.06 | 0.625 | 0.63 | 0.552 | |
| KermutStruc-GPModel Class=Gaussian process, Prior Information=ESM-2 + ProteinMPNN, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.622 | 0.625 | 0.564 | |
| KermutStruc-GP2026.06 | 0.622 | 0.625 | 0.564 | |
| MLP-SaProt-LastLayerBackbone=SaProt, Pooling=LastLayer2026.06 | 0.608 | 0.629 | 0.536 | |
| MLP-ESM2-LastLayerModel Class=Neural network, Prior Information=ESM-2, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.607 | 0.627 | 0.529 | |
| MLP-ESM2-LastLayerBackbone=ESM2, Pooling=LastLayer2026.06 | 0.607 | 0.627 | 0.529 | |
| Ridge-ESM2Model Class=Neural network, Prior Information=ESM-2, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.606 | 0.617 | 0.558 | |
| Ridge-ESM2-8MBackbone=ESM2-8M, Classifier=Ridge2026.06 | 0.606 | 0.617 | 0.558 | |
| MLP-ESM2-8MBackbone=ESM2-8M2026.06 | 0.597 | 0.618 | 0.538 | |
| MLP-OH-ESM2-650M-AugFeatures=One-Hot + ESM2-650M-Aug2026.06 | 0.581 | 0.605 | 0.548 | |
| Ridge-OH-ESM2-650M-AugFeatures=One-Hot + ESM2-650M-Aug, Classifier=Ridge2026.06 | 0.568 | 0.57 | 0.582 | |
| MLP-ESM2-8M-MeanPoolBackbone=ESM2-8M, Pooling=MeanPool2026.06 | 0.561 | 0.579 | 0.594 | |
| MLP-ESM2-8M-RandInitBackbone=ESM2-8M, Initialization=Random2026.06 | 0.558 | 0.575 | 0.572 | |
| MLP-OHBackbone=One-Hot2026.06 | 0.553 | 0.58 | 0.565 | |
| MLP-ESM2-650MBackbone=ESM2-650M2026.06 | 0.552 | 0.563 | 0.652 | |
| Ridge-ESM2-8M-MeanPoolBackbone=ESM2-8M, Pooling=MeanPool, Classifier=Ridge2026.06 | 0.552 | 0.55 | 0.603 | |
| SigGLM-OHModel Class=Linear + global non-linearity, Prior Information=—, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.55 | 0.569 | 0.57 | |
| SigGLM-OHBackbone=One-Hot2026.06 | 0.55 | 0.569 | 0.57 | |
| Ridge-OHModel Class=Linear, Prior Information=—, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.547 | 0.553 | 0.59 | |
| Ridge-OHBackbone=One-Hot, Classifier=Ridge2026.06 | 0.547 | 0.553 | 0.59 | |
| KermutSeq-GPModel Class=Gaussian process, Prior Information=ESM-2, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.541 | 0.539 | 0.615 | |
| KermutSeq-GP2026.06 | 0.541 | 0.539 | 0.615 | |
| Tanimoto-GPModel Class=Gaussian process, Prior Information=BLOSUM, Number of training data points=48, Evaluation regime=Cross-validation2026.06 | 0.52 | 0.517 | 0.588 | |
| Tanimoto-GP2026.06 | 0.52 | 0.517 | 0.588 | |
| BLOSUM50-ZeroShotMode=ZeroShot2026.06 | 0.396 | 0.386 | — | |
| ESM2-650M-ZeroShotMode=ZeroShot2026.06 | -0.044 | -0.04 | — |