This paper presents a novel method for predicting protein-protein interactions (PPIs) using large-scale language models (LLMs). To address the uncertainty inherent in conventional LLMs, which play a crucial role in understanding cellular mechanisms in complex diseases such as neurodegenerative diseases, metabolic syndrome, and cancer, we integrate fine-tuned LLaMA-3 and BioMedGPT models with a LoRA ensemble and Bayesian LoRA models to perform uncertainty quantification (UQ). We achieve competitive performance for PPI identification across diverse disease contexts, and by addressing model uncertainty, we enhance reliability and reproducibility in computational biology. This demonstrates our potential to contribute to the advancement of precision medicine and biomedical research.
