SAM-DTA

SAM-DTA predicts drug-target binding affinity using a sequence-agnostic modeling paradigm that characterizes proteins by their ligand interactions rather than by protein sequence.

Key Features:

• Sequence-Agnostic Approach: Proteins are characterized solely by their interactions with ligands, bypassing protein-sequence-based inputs.
• Multi-Head Training: Different proteins are treated as separate heads that are trained jointly to learn shared ligand representations.
• Superior Performance Metrics: Empirical evaluations report Mean Squared Error (MSE) 0.4261 versus 0.7612 and R-Square 0.7984 versus 0.6570 compared to DeepAffinity.
• Transfer Learning Capability: The model generalizes to unseen proteins after initial training, enabling prediction for novel targets.
• Cross-Dataset Evaluation: Validated through cross-dataset evaluations for prospective study scenarios.

Scientific Applications:

• Drug discovery: Predicting drug-target binding affinity to prioritize and identify potential therapeutic candidates.
• Analysis of novel or sequence-limited proteins: Enabling affinity prediction for proteins or datasets lacking extensive sequence information.

Methodology:

Models are trained on ligand interaction data using a multi-head learning framework where each head corresponds to a different protein while sharing underlying ligand representations, and proteins are represented by their ligand interactions rather than sequences.