Risk Phenotyping Before Graft Implantation: FTIR Spectroscopy and Machine Learning for Complementary Risk...

Area of Science:

• Biochemistry
• Immunology
• Data Science

Background:

• Kidney transplant rejection remains a significant obstacle to long-term graft survival.
• Current pre-transplant risk stratification methods for kidney recipients are insufficient.
• Identifying recipients at higher risk of rejection is crucial for improving outcomes.

Purpose of the Study:

• To evaluate the potential of pre-transplant serum Fourier-transform infrared (FTIR) spectra, analyzed by machine learning, in identifying kidney transplant recipients at increased risk of biopsy-proven rejection.
• To explore the utility of FTIR spectroscopy as a tool for pre-transplant risk stratification in kidney transplantation.

Main Methods:

• Retrospective analysis of 79 pre-transplant serum samples from kidney transplant recipients.
• Acquisition of FTIR spectra in the 600-1900 cm^-1 and 2800-3400 cm^-1 regions.
• Application of machine learning, including Naïve Bayes classifiers with Leave-One-Out Cross-Validation and Fast Correlation-Based Filter feature selection, with various preprocessing techniques.

Main Results:

• A leakage-aware nested machine learning model, utilizing second derivative transformation and normalization, achieved an AUC of 0.837.
• The best-performing model demonstrated an accuracy of 0.747 and a specificity of 0.821.
• Permutation testing confirmed the model's performance was significantly above chance (empirical p = 0.000999).

Conclusions:

• Pre-transplant serum FTIR spectroscopy, coupled with nested machine learning, successfully identified a spectral signal associated with subsequent biopsy-proven rejection.
• FTIR spectroscopy shows promise as a complementary method for pre-transplant biochemical risk phenotyping.
• External, multicenter validation is necessary before clinical implementation of this technique.