Table 3 The impact of molecular representation learning pre-training with different self-supervised task combinations

From: DTIAM: a unified framework for predicting drug-target interactions, binding affinities and drug mechanisms

MLM

MDP

MFGP

Warm Start

Drug Cold Start

Target Cold Start

✓

× 

× 

0.697

0.247

0.543

× 

✓

× 

0.586

0.203

0.509

× 

× 

✓

0.783

0.305

0.594

✓

✓

× 

0.82

0.418

0.673

✓

× 

✓

0.902

0.527

0.802

× 

✓

✓

0.851

0.489

0.769

✓

✓

✓

0.931

0.568

0.865

  1. Note: MLM, MDP, and MFGP are three self-supervised tasks for molecular pre-training, representing Masked Language Modeling, Molecular Descriptor Prediction and Molecular Functional Group Prediction, respectively. The ablation experiment is performed on the Yamanishi 08’s dataset under three experimental settings, and the evaluation metric is AUPR. Bold numbers represent the best performance, higher values are better.
Back to article page