MPAExpo-LM: Fine-Tuned Large Language Model for Mycophenolic Acid Exposure Estimation After Renal Transplantation

Area of Science:

• Pharmacokinetics and Pharmacogenomics
• Artificial Intelligence in Medicine
• Transplant Immunology

Background:

• Accurate mycophenolic acid (MPA) exposure monitoring is crucial for optimizing immunosuppressive therapy post-renal transplantation.
• Mycophenolate mofetil (MMF) and enteric-coated mycophenolate sodium (MPS) require precise MPA level management to prevent acute rejection.
• Existing methods for MPA exposure estimation may lack accuracy and efficiency.

Purpose of the Study:

• To develop and validate a novel large language model (LLM) for accurate MPA exposure estimation.
• To improve the precision of immunosuppressive therapy in renal transplant recipients.
• To offer a new computational approach for personalized medicine in transplantation.

Main Methods:

• Development of MPAExpo-LM, a fine-tuned LLM incorporating a differential attention denoising module.
• Training and testing on real-world renal transplant patient datasets.
• Validation using multi-center data and reinforcement fine-tuning for enhanced generalization.

Main Results:

• MPAExpo-LM achieved superior accuracy in MPA exposure estimation compared to existing methods.
• Root mean square errors (RMSE) of 8.78 mg·h/L (3 points) and 7.43 mg·h/L (4 points) on initial datasets.
• Demonstrated robust generalization across multiple hospitals with RMSEs of 7.342 mg·h/L (3 points) and 7.281 mg·h/L (4 points).

Conclusions:

• MPAExpo-LM offers a promising new pathway for precise immunosuppressive therapy in renal transplant patients.
• The model's strong generalization capability supports its potential for widespread clinical application.
• Accurate MPA exposure estimation using AI can significantly contribute to reducing acute rejection rates.