Model Parameters and Test Files for T-ALPHA: A Hierarchical Transformer-Based Deep Neural Network for Protein-Ligand Binding Affinity Prediction With Uncertainty-Aware Self-Learning for Protein-Specific Alignment

This dataset contains all the model parameters and test files required for reproducing the results presented in the paper:

T-ALPHA: A Hierarchical Transformer-Based Deep Neural Network for Protein-Ligand Binding Affinity Prediction with Uncertainty-Aware Self-Learning for Protein-Specific Alignment.

T-ALPHA is a novel deep learning model designed to predict protein-ligand binding affinity with state-of-the-art accuracy, integrating multimodal feature representations from three distinct channels—protein, ligand, and protein-ligand complex. This dataset includes:

1. Model Parameters: Fully trained model weights saved during the training of the T-ALPHA architecture. These parameters are essential for inference and validation of the model's performance.
2. Test Files: Protein-ligand complex datasets used for evaluation, including CASF 2016, LP-PDBbind, BDB2020+, and protein-specific test sets for SARS-CoV-2 main protease (Mpro) and the epidermal growth factor receptor (EGFR). The data is processed and formatted for direct use with T-ALPHA.

These files facilitate full reproducibility of the experiments, including evaluation benchmarks, uncertainty-aware self-learning for protein-specific alignment, and generalization performance on predicted structures.