A progressive fine-tuning framework with dynamic parameter selection for low-resource peptide-GPCR interaction...

Area of Science:

• Pharmacology and Drug Discovery
• Computational Biology
• Biochemistry

Background:

• G protein-coupled receptors (GPCRs) are critical drug targets.
• Peptide therapeutics are a growing area, but predicting their interactions with GPCRs (PepGI) is difficult.
• Existing drug-target interaction (DTI) models struggle with peptide data due to training on small molecules.

Purpose of the Study:

• To develop an effective computational framework for predicting peptide-GPCR interactions (PepGI).
• To address the challenges of data scarcity and poor generalization in current PepGI models.
• To create a transferable DTI modeling framework for cross-structure predictions.

Main Methods:

• Introduced a progressive fine-tuning framework with dynamic parameter selection using Fisher information.
• Pretrained on a large small molecule-GPCR dataset.
• Performed intermediate fine-tuning on peptide-target data and task-specific fine-tuning for low-resource PepGI scenarios.

Main Results:

• The proposed method significantly outperformed existing baseline models across multiple evaluation metrics.
• Demonstrated robust generalization capabilities in few-shot and cold-start settings.
• Successfully addressed the representation mismatch between small molecules and peptides.

Conclusions:

• The framework provides an effective solution for low-resource peptide-GPCR prediction.
• The approach offers a transferable strategy for drug-target interaction modeling across different ligand types.
• Advances computational prediction for peptide-GPCR drug discovery.