GeneralizedDTA

GeneralizedDTA predicts drug-target binding affinity by combining self-supervised pre-training on protein amino acid sequences and drug molecular graphs with multi-task learning and a dual adaptation mechanism to improve generalization to unknown drugs and out-of-distribution compounds.

Key Features:

• Pre-training for structural information: Employs self-supervised pre-training tasks on amino acid sequences and molecular graphs to capture detailed structural information of proteins and drug compounds.
• Mitigation of encoding variance: Reduces high variance associated with deep neural network-based encoding by enriching learned feature representations via pre-training.
• Multi-task learning framework: Uses a multi-task learning framework with a dual adaptation mechanism to narrow the task gap between pre-training and DTA prediction and to prevent overfitting.
• Handling out-of-distribution problems: Addresses out-of-distribution issues to improve predictive performance on novel or unrepresented drugs in labeled datasets.

Scientific Applications:

• Drug discovery: Predicts binding affinities for unknown drugs to accelerate identification of potential therapeutic candidates.
• Model generalization: Enhances generalization across diverse datasets to enable DTA prediction with limited labeled data.

Methodology:

Performs self-supervised pre-training on large-scale unlabeled data of amino acid sequences and molecular graphs, followed by a multi-task learning phase with a dual adaptation mechanism to adapt pre-trained representations for DTA prediction.