PLMAM-PLA: A Method Using Pretrained Language Models and Attention Mechanisms for Protein-Ligand Binding Affinity...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Protein-ligand binding affinity is critical for drug discovery and enzyme kinetics.
• Current deep learning models often require complex structural data.
• Sequence-based prediction offers a more practical and efficient approach.

Purpose of the Study:

• To develop a novel sequence-based deep learning model for predicting protein-ligand binding affinity.
• To improve the efficiency and accessibility of binding affinity prediction methods.
• To leverage pretrained language models for enhanced feature extraction.

Main Methods:

• Developed PLMAM-PLA, a sequence-based deep learning model.
• Utilized pretrained language models (ESM-2, MolFormer) for feature extraction from protein sequences and ligand SMILES.
• Employed dilated convolutional neural networks, SKNets, SENets, and attention mechanisms for feature enhancement and fusion.

Main Results:

• PLMAM-PLA effectively predicts protein-ligand binding affinity using only sequence and SMILES data.
• Ablation studies confirmed the contribution of individual model components.
• Visualization experiments demonstrated effective feature representation capture.
• Case studies showed strong generalization and superior performance compared to state-of-the-art methods.

Conclusions:

• PLMAM-PLA provides a powerful and efficient tool for sequence-based protein-ligand binding affinity prediction.
• The model demonstrates superior performance and generalization capabilities.
• This approach simplifies binding affinity prediction for drug discovery and related fields.